JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 114, B10312, doi:10.1029/2008JB006200, 2009 Click Here for Full Article

Mantle flow in subduction systems: The subslab flow field and implications for mantle dynamics Maureen D. Long1,2 and Paul G. Silver1,3 Received 6 November 2008; revised 8 June 2009; accepted 7 July 2009; published 28 October 2009.

[1] The character of the mantle flow field in subduction zones remains poorly understood, despite its importance for our understanding of subduction dynamics. In particular, little attention has been paid to mantle flow beneath subducting slabs. In order to identify processes that make first-order contributions to the global pattern of subslab mantle flow, we have compiled shear wave splitting measurements from subduction zones worldwide from previously published studies and estimated average splitting parameters for the subslab region. Globally, the subslab region is overwhelmingly dominated by trench-parallel fast splitting directions. We tested for relationships between splitting delay time, a measure of the strength of anisotropy, and parameters that are indicators of tectonic processes, such as trench migration velocity, convergence velocity and obliquity, age of subducting lithosphere, and slab dip, curvature, seismicity, and thickness. We used several different plate motion models to describe plate and trench motions and evaluated the differences among the models. We find that only one parameter, namely, the trench migration velocity in a Pacific hot spot reference frame, appears to correlate well with the strength of the subslab splitting signal. This finding supports a model in which the mantle beneath subducting slabs is dominated by three-dimensional flow induced by trench migration. We explore several implications of our model for various aspects of mantle dynamics, including the choice of a suitable reference frame(s) for mantle flow, the existence of a thin decoupling zone between slabs and the subslab mantle, and consequences for mass transfer between the upper and lower mantle. Citation: Long, M. D., and P. G. Silver (2009), Mantle flow in subduction systems: The subslab flow field and implications for mantle dynamics, J. Geophys. Res., 114, B10312, doi:10.1029/2008JB006200.

1. Introduction that drive it have implications for our understanding of the formation and migration of melt, volcanic production and [2] Subduction zones, where cold, negatively buoyant hazard, the tectonics of the fore-arc and back-arc region, lithosphere is recycled into the Earth’s mantle, represent and the style of mantle convection and mixing. In particular, both key components of the plate tectonic system and prime the characterization of mantle flow beneath downgoing sites of volcanic and seismological hazard [e.g., Stern, slabs, and the associated implications for how slabs interact 2002]. Subduction zones represent perhaps the most tecton- with surrounding mantle material, has received little atten- ically complicated regions on the Earth, and although much tion. In this study, we focus on anisotropy and mantle flow progress has been made in understanding their structure and in the subslab region, for several reasons. First, despite its dynamics, several crucial aspects of the subduction process importance for a complete understanding of the mantle flow remain enigmatic. In particular, the dynamic interaction field associated with subduction, the subslab region has between sinking slabs and the surrounding mantle flow been paid comparatively little attention in studies of mantle field remains poorly understood. The simplest model for flow in subduction systems, in contrast to the mantle wedge. this flow field invokes two-dimensional flow, with corner Second, as discussed later in this paper, the relationship flow in the mantle wedge above the slab and entrained flow between seismic anisotropy and mantle flow is much more beneath it; however, this model is increasingly contradicted straightforward for the subslab region than for the wedge, by seismological (and other) observations. The character of where unusual olivine fabrics [Jung and Karato, 2001] or the subduction zone flow field and the nature of the forces effects from melt [Holtzman et al., 2003] are possible and the interpretation of seismic anisotropy can be more am- 1Department of Terrestrial Magnetism, Carnegie Institution of biguous. Third, as we will show, observations of anisotropy Washington, Washington, D. C., USA. in the subslab region are much simpler and exhibit much 2Now at Department of Geology and Geophysics, Yale University, New Haven, Connecticut, USA. less regional variation than do comparable observations for 3Deceased 7 August 2009. the wedge region. [3] The most direct constraints on the geometry of mantle Copyright 2009 by the American Geophysical Union. flow are obtained through observations of seismic anisot- 0148-0227/09/2008JB006200$09.00 ropy. In particular, the observation that seismic shear waves

B10312 1of25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312

successfully explains many aspects of global subduction zone splitting patterns. This model, in which the subduction zone flow field is controlled by 3-D flow induced by the migration of the trench with respect to the underlying mantle, explains the preponderance of trench-parallel fast directions beneath the slab as well as the observed correla- tion between trench migration velocity and delay times for both the subslab and wedge regions that we observed in a global compilation of splitting measurements. [6] Here, we expand our previous work and present a more complete investigation of the relationships between splitting parameters and parameters that describe the sub- duction process, with a focus on the subslab mantle flow Figure 1. Sketch of various raypaths involved in isolating field. The work presented here extends our previous inves- slab versus subslab anisotropy, including local S phases tigation in four ways. First, we have augmented our global from slab earthquakes (white stars), teleseismic SKS-type database of subduction zone splitting measurements with phases measured at stations (black squares) above the slab, characterizations of source-side anisotropy from teleseismic and direct S from slab earthquakes measured at teleseismic measurements of intermediate-depth earthquakes occurring distances. in subduction zones [e.g., Wookey et al., 2005; Rokosky et al., 2006; Mu¨ller et al., 2008]. These data provide a more are birefringent, and undergo ‘‘splitting’’ as they propagate direct constraint on anisotropy and flow beneath downgoing through anisotropic regions of the Earth, has provided a tool slabs, as the measurements are uncontaminated by anisotropy for studying the local geometry of mantle flow beneath in the wedge. Second, we present comparisons between seismic stations (see overviews by Silver [1996], Savage subslab splitting parameters and a wide variety of parame- [1999], and Park and Levin [2002]). Shear wave splitting ters that describe subduction, such as trench migration studies, which measure the fast direction f and the delay velocity, convergence velocity and obliquity, age and spread- time dt for each shear arrival, have increased in popularity ing history of subducting lithosphere, slab dip, curvature, and now represent a standard tool for probing mantle seismicity, thickness, and morphology, and arc length. deformation. Subduction zones have been a popular target These comparisons allow us to test alternative hypotheses for shear wave splitting practitioners, and since early studies concerning factors that control anisotropy beneath the by Ando et al. [1983], Fukao [1984], and Bowman and mantle wedge. Third, we revisit the comparisons between Ando [1987], several dozen splitting studies in subduction subslab split times and trench migration velocities shown by zone regions have been performed. With the wealth of such Long and Silver [2008] using a variety of global reference data now available, it is both important and timely to frames and plate motion models [Schellart et al., 2008; undertake comprehensive studies of shear wave splitting Lallemand et al., 2008]; in principle, this should allow for behavior in subduction zones globally, with the aim of the identification of an ideal ‘‘rest frame’’ for the convecting determining the dominant properties of the mantle flow mantle, at least in the Pacific, around which most subduc- field associated with the subduction process. tion zones in our compilation are located. Finally, we [4] Although shear wave splitting in subduction zone address in detail some of the implications of the Long and regions has been extensively characterized, consensus on Silver [2008] model for mantle dynamics, including the how it relates to mantle processes has not been forthcoming, nature of a likely decoupling zone beneath subducting slabs for several reasons. First, the inherent structural complexity and constraints on mass exchange between the upper and of subduction zones (which comprise the overriding plate, lower mantle. the superslab mantle wedge, the slab itself, and the subslab mantle) not only makes the interpretation of splitting 2. A Global Data Set of Subslab Shear Wave measurements difficult, but it makes the measurements Splitting Parameters in Subduction Zones themselves challenging; complex anisotropic structure 2.1. Methods for Isolating the Subwedge Contributions beneath a station leads to complicated splitting patterns which exhibit significant variations [e.g., Long and van [7] Isolating the contribution to splitting observed at the der Hilst, 2005] and good data coverage is needed to fully surface from anisotropy beneath subducting slabs is not characterize such complexity. Second, the wide variety of completely straightforward; here we describe the approach splitting behaviors observed in subduction zones has posed taken by Long and Silver [2008] and also applied in this challenges for interpretation: large variations in both f and study. Teleseismic shear phases such as SKS, SKKS, and dt have been observed, significant lateral heterogeneity is direct S from deep events are commonly used in splitting often present, and the preponderance of fast directions that studies to characterize upper mantle anisotropy beneath the are parallel to the trench are contrary to the expectation of station. In the case of a subduction zone, however, such simple two-dimensional (2-D) entrained flow. phases sample several components of the subduction sys- tem, including the subslab mantle, the slab itself, the mantle [5] A promising approach to this problem lies in a systematic study of splitting constraints in subduction zone wedge, and the overriding plate (Figure 1). In contrast, regions and a search for relationships between splitting direct S waves from earthquakes originating in the subduct- parameters and parameters that describe subduction. We ing slab recorded on the overriding plate will only be recently proposed a model [Long and Silver, 2008] that affected by anisotropy in the mantle wedge and in the

2of25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312 overriding plate (and perhaps a small portion of the slab and van der Hilst, 2006; Wirth and Long, 2008] and since itself). If splitting from both local S phases and teleseismic the frequency bands used in local S splitting studies often S or SK(K)S phases is measured, therefore, and if a careful vary from study to study and are generally higher than those correction for splitting in the wedge and overriding plate is used in studies of teleseismic splitting, a direct comparison applied to teleseismic measurements is applied, then an among studies is difficult. Despite these challenges, how- estimate may be obtained for the anisotropy beneath the ever, our method of accounting for contributions to tele- mantle wedge. This is the approach we have taken to obtain seismic splitting from the mantle wedge and overriding estimates of the subwedge splitting signal; where possible, plate and attributing the rest of the splitting signal to we have focused on stations located close to the trench, anisotropy in the subslab mantle and the slab itself has where teleseismic phases sample a large volume of subslab proven successful at providing a rough estimate of subslab mantle material and little, if any, of the mantle wedge. This splitting, although the errors on individual estimates may be approach can isolate the contribution from the subwedge large [Long and Silver, 2008]. mantle, but an ambiguity remains as to the relative contri- [10] Estimates of source-side anisotropy from splitting butions to this signal from the downgoing slab itself versus measurements on teleseismic phases from earthquakes that the subslab mantle. As we discuss in section 2, several lines originate in subducting slabs (Figure 1) provide another of evidence argue for a primary contribution to the sub- type of data that is useful for characterizing subwedge wedge splitting signal from active mantle flow beneath the anisotropy [Russo and Silver, 1994]. In this approach, slab; slab anisotropy does not appear to dominate subwedge splitting due to anisotropy beneath the source, splitting splitting on a global scale. For now, however, we restrict our due to upper mantle anisotropy near the receiver, and (in analysis to determining the subwedge splitting signal with- some cases) splitting due to anisotropy in the lowermost out distinguishing between anisotropy in the slab versus the mantle are characterized separately; detailed descriptions of subslab mantle. the methods used to isolate each contribution are given by [8] The correction of teleseismic splitting measurements Wookey et al. [2005], Rokosky et al. [2006], and Mu¨ller et to remove the effects of wedge anisotropy is only straight- al. [2008]. This type of data was not included in our original forward for the simplified case where the layers of anisot- compilation; its inclusion here has allowed us to add new ropy (in the wedge, the subslab mantle, and perhaps in the subduction zones (the Marianas and Scotia) to our compi- slab itself) are oriented either parallel or perpendicular to lation of subwedge anisotropy. In many ways, this method each other. For this geometric configuration, the delay times provides a more direct constraint on subwedge anisotropy, can simply be added (for layers with the same orientation) since the mantle wedge and overriding plate of the subduc- or subtracted (for layers with a perpendicular orientation). In tion system are not sampled by this type of measurement practice, however, this correction is usually more difficult, (although, of course, anisotropy on the receiver side and in for several reasons. First, splitting patterns in the mantle the lowermost mantle must be properly accounted for). The wedge are often very complicated [e.g., Pozgay et al., 2007; characterization of source-side splitting from slab earth- Wiens et al., 2008; Long and Silver, 2008] and the inferred quakes, along with the use of unconventional raypath anisotropy is often complex and changes on short length configurations such as the use of sS phases to study scales; in such cases, care is needed to choose appropriate anisotropy in the wedge near the source [e.g., Anglin and local S raypaths to correct the corresponding teleseismic Fouch, 2005], has the potential to add further to our global phases. Second, for the case where the measured fast direc- compilation in the future. tions for teleseismic and local are not either parallel or [11] A key consideration when assigning average delay perpendicular to each other, correcting the teleseismic split- times to the subwedge portion of the subduction system is ting for the contribution from wedge anisotropy is more accounting for potential differences in slab and subslab path complicated than simply subtracting (or adding) the wedge lengths among different regions. These path lengths are split time from the teleseismic split time. Ideally, in this case, controlled by the positions of the seismic stations under the back azimuthal variations in teleseismic splitting should study relative to the trench and by the dip and thickness of be carefully characterized and forward modeling of multiple- the subducting slab. We have preferentially chosen stations layer anisotropy [Silver and Savage, 1994] should be carried located close to the trench in order to minimize these out to correctly identify the splitting parameters associated effects, but some differences in slab and subslab path with the lower (subslab) anisotropic layer. Fortunately, how- lengths exist. Because the depth of the bottom of the ever, in many cases a simple addition or subtraction of split anisotropic layer beneath subducting slabs is poorly known times can be carried out, as measured fast directions for and may vary in different regions, it is difficult to explicitly teleseismic phases at stations tend to be (mostly) trench- correct subwedge delay times for the effect of path length. parallel or (in a few cases) trench-perpendicular, and many We do, however, estimate the range of subwedge path wedges display a characteristic transition from trench- lengths associated with each range of delay time estimates, parallel fast directions to trench-perpendicular fast direc- as discussed below. tions as well. 2.2. Description of Regional Studies and Results [9] A third challenge to correctly accounting for the wedge contribution to teleseismic phases is a possible [12] Our compilation is based on over two dozen previ- dependence of measured splitting parameters on the fre- ously published studies and consists of 22 individual quency content of the waves under study. Many studies estimates for subwedge shear wave splitting in 15 different have shown that shear wave splitting of local S phases in subduction zones, as described below. For subduction subduction zones can be frequency-dependent [Fouch and systems with long arc lengths where data are available Fischer, 1996; Marson-Pidgeon and Savage, 1997; Long for stations located at different locations along the arc,

3of25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312

Figure 2. Summary sketch of subslab splitting worldwide. Locations of trenches from the Gudmundsson and Sambridge [1998] slab model are shown, along with an arrow indicating the approximate dominant fast splitting direction. Nearly all subduction systems, with the major exception of Cascadia, exhibit primarily trench-parallel fast directions. Some systems (Japan, Sumatra) exhibit both trench-parallel and oblique f, while South America exhibits mainly trench-parallel f with a localized region of trench-perpendicular fast directions. Note that the arrows on this map provide a summary of the dominant fast directions in any given region but do not correspond one-to-one to the along-strike segments that are included in the compilation. Some regions (e.g., Japan) have a mix of fast directions that are not included on this sketch for simplicity. For a more detailed discussion of splitting patterns observed in any given subduction zone, see section 2.2. such as South America or the Aleutians, we have made included in the original Long and Silver [2008] compilation separate estimates of subwedge splitting (and other tecton- because of its proximity to the strike-slip Alpine Fault and ic parameters, described in section 4) for individual seg- the associated complicated tectonic setting, but here we ments of the arc. A map that summarizes our estimates for have chosen to use results from the northeastern part of the subwedge splitting for subduction zones worldwide is North Island, which is located closest to the Hikurangi shown in Figure 2. Error bars are estimated by using the trench and furthest from the Alpine Fault. Comparison of range of split times reported by authors of each study; the SKS splitting with local S measurements at low frequencies error bars are, necessarily, fairly approximate, but they are [Audoine et al., 2004; see also Greve et al., 2008] yields an designed to encompass most of the range of subwedge delay estimate for subwedge splitting beneath Hikurangi of dt = times allowed by the data and roughly represent a 95% 1.5 ± 0.5 s, with a trench-parallel fast direction. The confidence region [see also Long and Silver, 2008]. The northeasternmost stations used by Audoine et al. [2004] average subwedge delay time estimates, error bars, range of are located 20–40 km above the Hikurangi slab, so path lengths, and data sources are summarized in Table 1. subwedge upper mantle path lengths are roughly 360 km. [13] Shear wave splitting in the northern part of the [14] For the Sumatra/Indonesia subduction zone, we Tonga-Kermadec subduction zone has been studied exten- relied on measurements reported by Hammond et al. sively, although most studies have focused on measuring [2006], as well as measurements from permanent station local S splitting [Fischer and Wiens, 1996; Fischer et al., PSI reported by Long and Silver [2008]. Hammond et al. 1998; Smith et al., 2001]. SKS splitting was measured at a [2006] reported a nearly isotropic wedge, with a small few stations in northern Tonga by Fischer et al. [1998], but contribution to local S splitting from anisotropy in the the stations were located fairly far away from the trench and overriding lithosphere, and significant SKS splitting with the measurements likely sample the wedge much more than a fast direction oblique to the trench that they attribute to the subwedge mantle [see also Fischer et al., 2000]. For our anisotropy within the slab. Our own local and teleseismic compilation of subwedge anisotropy, therefore, we relied on measurements at station PSI suggest that between 0.8 s and measurements for permanent station RAO documented by 1.4 s of subwedge splitting contributes to the teleseismic Long and Silver [2008]. A comparison of SKS and local S splitting signal, with little or no splitting due to wedge splitting at this station reveals a trench-parallel fast direction anisotropy. The subwedge fast splitting direction is close to and a delay time of 1.7 ± 0.7 s that can be attributed to trench-parallel. Station PSI is located 200 km above the anisotropy beneath the slab. Station RAO is located approx- subducting Indonesia slab, so subwedge path lengths in the imately 100 km above the subducting Tonga slab, so the upper mantle beneath this station are 200 km. subwedge upper mantle path lengths are roughly 300 km. [15] The tightest constraints on subwedge anisotropy Further to the south, shear wave splitting in the Hikurangi beneath the Mariana subduction zone come from source- subduction zone off the east cost of the North Island of New side splitting for intermediate depth earthquakes, studied by Zealand has been extensively documented [e.g., Gledhill Wookey et al. [2005]. This work yielded split times that and Gubbins, 1996; Brisbourne et al., 1999; Matcham et range from 0.5 to 1.0 s for the Marianas, with consistently al., 2000; Audoine et al., 2004; Marson-Pidgeon and trench-parallel fast directions, but this represents measure- Savage, 2004; Morley et al., 2006]. This region was not ments from only two earthquakes. These events originated at

4of25 B10312

Table 1. Summary of the Constraints on Subwedge Splittinga FLOW MANTLE ZONE SUBDUCTION SILVER: AND LONG Subwedge Path Number of Lengths in Upper Stations Subduction Zone Type of Data Subwedge dt (s) Subwedge f Mantle (km) (Events) Source Tonga-Kermadec SKS, local S 1.0–2.4 Trench-parallel 300 1 Long and Silver [2008] Hikurangi SKS, local S 1.0–2.0 Trench-parallel 360 2 Audoine et al. [2004] Sumatra/Indonesia SKS, local S 0.8–1.4 Trench-parallel, oblique 200b 20 Long and Silver [2008] and Hammond et al. [2006] Izu-Bonin SKS 0.5–1.0 Trench-parallel, some oblique; 300 4 Wirth and Long [2008] splitting patterns complex Marianas Source-side S 0.5–1.0 Trench-parallel 200 (2) Wookey et al. [2005] Ryukyu SKS, local S 0.0–0.2 Trench-parallel, if any 280–350 8 Long and van der Hilst [2005, 2006] 5of25 Japan (N. Honshu, SKS, local S 0.2–0.8 Oblique, some trench-parallel; 200–320 20 Long and van der Hilst [2005], Nakajima and Hokkaido) splitting patterns complex Hasegawa [2004], and Wirth and Long [2008] Kamchatka SKS, local S 0.4–0.8 Trench-parallel 200–340 5 Peyton et al. [2001] and Levin et al. [2004] Aleutians SKS, local S 0.0–0.3 Trench-parallel, if any 300 3 Long and Silver [2008] Cascadia SKS 1.0–1.5 Trench-perpendicular 300–350 15 Currie et al. [2004] Caribbean SKS, limited local S 1.0–2.0 Trench-parallel 300 7 Pin˜ero- Feliciangeli and Kendall [2008] Calabria SKS 1.0–2.0 Trench-parallel 200–350 30 Baccheschi et al. [2007] Middle America SKS, local S 1.0–1.7 Trench-parallel 250–360 20 Abt et al. [2005, 2009] South America SKS, local S, source-side S 0.5–1.5 Variable, but mostly trench-parallel 0–200 (source-side), 20 (10) Rokosky et al. [2006] (source-side), Polet et al. [2000] 100–300 (SKS) (SKS, local S), Russo and Silver [1994] (all types) Scotia Source-side S 1.0–1.6 Trench-parallel 125–390 (20) Mu¨ller [2001] aFrom the published literature, as described in section 2.2. We have listed the most relevant citations, but additional relevant studies are discussed in the text. For each subduction zone, the range of subwedge delay times and fast directions allowed by the data are listed, along with the approximate number of stations (or events for source-side splitting) used to obtain the estimate. We estimate the subwedge upper mantle path length by assuming that the depth of the bottom of the anisotropic layer is at 400 km and taking into consideration the depth to the slab beneath the stations used (for SKS/local S studies) and/or the depth of the earthquakes (for source-side studies). bFor PSI; large range for other stations. B10312 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312 depths of 200 km in the slab, so the rays sample 200 km Nakajima and Hasegawa [2004] and Nakajima et al. of subslab upper mantle. The splitting of local S phases has [2006], although these studies examined energy in much been studied in the Marianas by Fouch and Fischer [1998], higher frequency bands than corresponding studies of tele- Volti et al. [2006], and Pozgay et al. [2007], but constraints seismic splitting, making direct comparisons difficult. These on SKS splitting are very sparse [Fouch and Fischer, 1998], workers documented small split times (0.1 s) and a largely due to the unfavorable distribution of seismicity in striking transition in fast direction from trench-parallel close the epicentral distance range used to study SKS. For the Izu- to the trench to trench-perpendicular in the back arc, which Bonin arc, there are published constraints on mantle wedge they attribute to B-type olivine fabric in the corner of the anisotropy from Anglin and Fouch [2005], but few con- mantle wedge. Wirth and Long [2008] recently measured straints on SKS. A preliminary investigation of SKS and local splitting in northern Honshu at frequencies compara- local S splitting at F-net stations located in the northern part ble to those used to measure teleseismic splitting [Long and of the Izu-Bonin arc by Wirth and Long [2008] yielded van der Hilst, 2005], and found split times of 0.3 s. Given generally trench-parallel SKS fast directions, but some SKS these observations, our best estimate for northern Japan is splitting patterns exhibit back azimuthal variations and the that 0.5 ± 0.3 s of splitting can be attributed to the subwedge interpretation of the subwedge splitting signal requires more region, with fast directions that range from trench-parallel to detailed forward modeling. The few local S measurements oblique to the trench. Raypath lengths used in this estimate are from deep events [Wirth and Long, 2008] and are not range from 200 to 320 km. well constrained; the local raypath distribution is not ideal [18] Further to the north, in Kamchatka, both teleseismic for isolating the subslab contribution to anisotropy. We and local splitting have been studied using data from the therefore assign 0.5–1.0 s of trench-parallel subslab Side Edge of the Kamchatka Slab experiment [Peyton et al., splitting to Izu-Bonin in our complication, but we empha- 2001; Levin et al., 2004]. Peyton et al. [2001] found size that this is based on preliminary results. significant (dt 1 s) trench-parallel SKS splitting at stations [16] Shear wave splitting at stations in the Ryukyu arc in southern Kamchatka with much smaller splitting (dt was studied by Long and van der Hilst [2005, 2006], who 0.1–0.3 s) from local S phases. Mantle wedge anisotropy carried out a comparison between teleseismic and local from local S splitting was studied in greater detail by Levin shear wave splitting in the same frequency bands and et al. [2004]; they found slightly higher delay times (dt concluded that most of the teleseismic splitting signal can 0.2–0.6 s) and complex spatial patterns, with a suggestion be attributed to the mantle wedge, with little or no splitting that fast directions exhibit a change from trench-perpendic- due to subwedge anisotropy required to explain the data. ular close to the trench to trench-parallel farther in the back Further modeling work [Long et al., 2007; Kneller et al., arc (the opposite trend was observed in Japan by Nakajima 2008] has confirmed this interpretation; in particular, Kneller and Hasegawa [2004]). Taken together, these studies sug- et al. [2008] found that a model with significant anisotropy gest that between 0.4 and 0.8 s of trench-parallel splitting in the mantle wedge due to B-type olivine fabric with a small can be attributed to the subwedge mantle beneath Kam- contribution to trench-parallel splitting (most likely less chatka. The stations used to obtain the SKS measurements than 0.2 s) from the slab or subslab mantle explains the of Peyton et al. [2001] were generally located between 60 data well. Therefore, in our compilation we attribute and 200 km above the subducting slab, which roughly between 0 and 0.2 s of trench-parallel splitting to the corresponds to subwedge upper mantle path lengths between subwedge mantle beneath Ryukyu. Ryukyu stations are 200 and 340 km. Our constraints on subwedge anisotropy located mainly on the arc islands and the subwedge raypaths for the Aleutian subduction zone come from measurements lengths range from approximately 280–340 km. made by Long and Silver [2008] for three permanent stations [17] Both teleseismic and local splitting beneath Japan in the Aleutian island arc (ATKA, NIKO, and SMY). A has been extensively studied [e.g., Fouch and Fischer, comparison of local and teleseismic splitting at these 1996; Sandvol and Ni, 1997; Fischer et al., 1998; Nakajima stations suggests that there is little, if any, splitting from and Hasegawa, 2004; Long and van der Hilst, 2005; the subwedge region. SKS splitting measurements reveal Nakajima et al., 2006; Wirth and Long, 2008] and the 0.5–1.0 s of (nearly) trench-parallel splitting, but a observed splitting patterns in Japan are perhaps the most comparison with local S splitting suggests that the anisot- complicated of any subduction zone in the world. This is ropy being sampled by SKS waves at these stations is in the particularly true at stations located near the Kanto region mantle wedge. Therefore, we estimate a contribution to near the triple junction of the Pacific, Philippine Sea, and splitting from the subwedge region of 0–0.3 s at Aleutian Eurasian plates, where the slab morphology (and presum- stations. ATKA, NIKO, and SMY are all located approxi- ably the resulting flow patterns) is complicated [e.g., Long mately 100 km above the subducting slab, which corre- and van der Hilst, 2005; Wirth and Long, 2008]. With the sponds to 300 km of subwedge upper mantle path length. caveat that the causes for the complex splitting patterns in [19] The Cascadia subduction zone in the northwestern Japan are not yet well understood, we have focused on United States is characterized by a dearth of seismicity stations in northern Honshu and on Hokkaido to obtain a beneath 80 km, so unlike most subduction zones world- general estimate for subwedge splitting beneath Japan. wide, direct comparisons between local and teleseismic Teleseismic splitting in this region is complex, although shear wave splitting are difficult. Shear wave splitting in stations closest to the trench in both Honshu and Hokkaido the Pacific Northwest is becoming increasingly well char- exhibit clear trench-parallel fast directions and split times of acterized, thanks to new data from the USArray component up to 0.8–1.0 s. Fast directions measured at other stations of Earthscope and other temporary deployments in the in the region are generally oblique to the trench. Local region, but only a few studies have examined splitting for splitting for northern Honshu and Hokkaido was studied by stations close to the trench that sample large volumes of

6of25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312 subwedge mantle. We have relied mainly on SKS measure- indicates that there is significant trench-parallel splitting ments made by Currie et al. [2004]; they found that stations beneath the subducting Middle America slab; we have in the Cascadia fore-arc regions exhibit 1.0–1.5 s of assigned a range of 1.0–1.7 s for dt in our compilation. splitting, with a fast direction that is roughly perpendicular We focused on results from TUCAN stations located less to the strike of the trench. There are few constraints on than 150 km above the subducting slab, which corresponds wedge splitting in this region; one early study by Cassidy to subwedge upper mantle path lengths between 250 and and Bostock [1996] found local split times of 0.3 for the 360 km. deepest earthquakes beneath Vancouver Island, but even the [22] Further to the south, in the South America subduc- deepest events in their data set may not sample much mantle tion zone, shear wave splitting has been investigated by wedge material. However, Currie et al. [2004] argue that several different groups using both permanent stations and most of the splitting they measure can be attributed to temporary deployments [Russo and Silver, 1994; Bock et anisotropy beneath the subducting slab, based on the large al., 1998; Polet et al., 2000; Anderson et al., 2004]. Using split times observed at stations close to the trench for data from stations located along the South American margin raypaths that mainly sample the subwedge region, and we and using a variety of raypath combinations, Russo and follow their interpretation. The fore-arc stations used by Silver [1994] inferred significant splitting beneath the sub- Currie et al. [2004] to estimate SKS splitting were located ducting Nazca slab, with mostly trench-parallel fast direc- 50–100 km above the subducting slab, corresponding to tions. Later studies [Bock et al., 1998; Polet et al., 2000] subwedge upper mantle path lengths between 300 and found a mix of trench-parallel, trench-perpendicular, and 350 km. Their observation that subwedge anisotropy beneath oblique SKS fast directions for South America; a compar- Cascadia exhibits a trench-perpendicular fast direction is ison of the large split times observed for SKS phases with highly unusual in the global data set. the generally small split times observed for local S phases [20] Constraints for the Caribbean subduction zone indicates that there is significant subwedge splitting beneath came from the work of Pin˜ero-Felicangeli and Kendall South America. In particular, Polet et al. [2000] identified a [2008], who measured both local and teleseismic splitting, localized region of trench-normal fast directions but con- although local measurements are somewhat sparse. Pin˜ero- cluded that elsewhere in South America, trench-parallel fast Felicangeli and Kendall [2008] argue for 1.0 s of trench- directions dominated the SKS signal. Using data from a parallel splitting due to subwedge anisotropy in the temporary deployment in Chile and Argentina, Anderson et Caribbean; we have assigned a large error bar to the al. [2004] concluded that there is a large contribution (1s estimate (due to the weak constraints on wedge anisotropy) or more) to the observed trench-parallel SKS splitting from and we use a range of 0.5 to 1.5 s in our compilation. We within or below the slab, with only a small (0.15 s on have focused on the stations located in the Caribbean arc average) contribution from the wedge and overlying plate itself, which are located approximately 100 km above the (M. Anderson, personal communication, 2009). Subwedge subducting slab. We take a similar approach to assigning an anisotropy beneath Chile has also been studied using estimate for subwedge splitting beneath the Calabrian arc in observations of source-side splitting for earthquakes origi- Italy; in this region, SKS splitting has been studied by nating in the South American slab by Rokosky et al. [2006]; several workers [Schmid et al., 2004; Civello and Margheriti, they identified consistently trench-parallel fast directions 2004; Baccheschi et al., 2007] but there are few constraints with subwedge split times between 0.5 and 1.6 s. All of on wedge anisotropy, so it is not entirely clear how much of these studies paint a fairly consistent picture of subwedge the SKS signal can be attributed to mantle flow beneath the splittingbeneathSouthAmerica,whichappearstobe wedge. Baccheschi et al. [2007] argue that most of the SKS dominated by trench-parallel fast directions with localized splitting in their data set is due to trench-parallel anisotropy regions of trench-perpendicular or oblique directions. Sub- beneath the slab, as the large split times (up to 3 s) are too wedge split times are on the order of 0.5–1.5 s, depending large to be attributed solely to anisotropy in the lithosphere on locality. Subwedge path lengths vary greatly depending or the mantle wedge We assign a subwedge splitting time of on the type of data; for source-side measurements, event 1.5 ± 0.5 s with a trench-parallel fast direction for the depths vary from 220 km to transition zone depths. We Calabrian arc in our compilation. The stations are located preferentially selected stations located closer to the trench for generally between 50 and 200 km above the subducting SKS and local S measurements, but also included measure- Calabria slab, which corresponds roughly to subwedge ments for a few stations located farther away; for this type of upper mantle path lengths of 200–350 km. data, the subwedge path lengths vary from 100 to 300 km. [21] Splitting constraints from the Middle America sub- [23] The South Sandwich subduction zone lies to the duction zone come from studies using data from the south of South America; shear wave splitting has been TUCAN experiment [Abt et al., 2005, 2009; Hoernle et measured at a few stations located in the arc itself [Mu¨ller, al., 2008]. Mantle wedge anisotropy in this region has been 2001] but the station coverage is extremely sparse and the well characterized, as the three-dimensional distribution of lack of high-quality SKS and local S measurements for seismic anisotropy was probed using a tomographic imag- stations located in the arc itself makes interpretation diffi- ing method for local S splitting measurements [Hoernle et cult. More direct constraints on subwedge splitting are al., 2008; Abt et al., 2009]. SKS splitting has been inves- provided by a recent study of source-side anisotropy using tigated at a subset of the TUCAN stations; consistently earthquakes in the downgoing slab and receivers located in trench-parallel SKS fast directions with large split times (up Antarctica [Mu¨ller et al., 2008]. This work demonstrated to 2.0–2.5 s) were documented by Abt et al. [2005]. A that significant trench-parallel splitting occurs beneath the comparison of local S split times in this region, which wedge, with split times generally between 0.4 and 1.6 s, generally range up to 0.3–0.5 s, with SKS measurements depending on earthquake depth. Measurements from shal-

7of25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312 lower earthquakes are most relevant for our work, since they slab itself does not contribute significantly to the observed have the longest paths lengths through the subwedge global subwedge splitting signal. This lack of a correlation mantle, and the split times from such events in the Mu¨ller argues against models that invoke slab anisotropy to et al. [2008] study were generally greater than 1s. explain global SKS splitting patterns, as discussed further Therefore, we assigned the South Sandwich subduction in section 8. Third, in regions where estimates of subwedge zone a range of subwedge split times from 1.0 to 1.6 s, splitting are available from multiple data sets (both from a with trench-parallel fast directions. The hypocentral depths comparison of local and teleseismic splitting and from used in the Mu¨ller et al. [2008] study ranged from 10 km source-side splitting from slab earthquakes), these estimates to 275 km, which corresponds to a range of subwedge path are very consistent, despite the fact that different raypath lengths in the upper 400 km of the Earth of 125–390 km. configurations sample the slab in different ways. Therefore, [24] Our compilation of subwedge splitting parameters we interpret the subwedge splitting signal as mainly repre- makes use of two distinctly different types of data; we senting anisotropy and mantle flow beneath the subducting exploit both SKS splitting estimates which have been slab. corrected for the effect of wedge anisotropy (inferred from local S splitting) and estimates of source-side splitting from 2.4. Interpreting Subslab Fast Directions: Mineral slab earthquakes measured at distant stations. The success- Physics Considerations ful comparison of results using these two different [26] Until very recently, the available mineral physics approaches in the same region provides a good compatibil- data have uniformly suggested that the subslab mantle is ity check and argues for the robustness of our SKS dominated by A-, C-, or E-type olivine and that the correction procedure. South America represents the best conditions needed for B-type olivine fabric (high stress, opportunity to check the two different types of measure- low temperature, and significant hydrogen content) are not ments against each other; the source-side splitting estimates present beneath the slab [Jung and Karato, 2001; Jung et from Russo and Silver [1994] and Rokosky et al. [2006] are al., 2006; Karato et al., 2008]. This would imply that fast very consistent with our estimates obtained from the SKS directions beneath the slab should be interpreted as being and local S splitting measurements of Russo and Silver roughly parallel to the local mantle flow direction. Recent [1994], Polet et al. [2000], and Anderson et al. [2004]. experiments, however, have provided some evidence for a Specifically, Rokosky et al. [2006] document subslab split- pressure-induced transition to B-type olivine fabric [Jung et ting of 1 s with fast directions that fall within 25° of the al., 2009]; taken at face value, these experiments suggest local trench strike at latitudes between 26°–28°S; just to the that the mantle flow direction would be orthogonal to the south, the northernmost stations of the CHARGE experi- fast direction for depths greater than 80–90 km. This ment exhibit trench-parallel SKS splitting with delay times prediction, however, is difficult to reconcile with existing between 0.5 s and 1.5 s [Anderson et al., 2004], while petrological and seismological evidence, which strongly local S phases exhibit much smaller splitting (0.15 s suggests that only limited regions of the upper mantle are (M. Anderson, personal communication, 2009)). Rokosky dominated by B-type olivine fabric [see Long, 2009, and et al. [2006] find similar subslab splitting at latitudes references therein]. For example, modeling studies that rely between 20°–24°S; again, this is generally consistent with on the A-type (or similar) fabric paradigm have been very SKS and local S splitting [Russo and Silver, 1994; Polet et successful in explaining anisotropy observed beneath ocean al., 2000], particularly for stations located close to the basins [e.g., Behn et al., 2004; Conrad et al., 2007; Becker, trench. As estimates of source-side anisotropy from slab 2008]; in particular, studies which have compared azimuthal earthquakes in a variety of regions become available, we anisotropy from surface wave observations to the predic- expect to perform additional comparisons between different tions from geodynamical models have shown that A-type types of data, but the existing estimates suggest that our (or similar) fabric likely predominates at asthenospheric SKS correction procedure correctly isolates the subwedge depths, contradicting the prediction of Jung et al. [2009] splitting contribution. that B-type fabric should be present below 90 km. The relationship between strain and anisotropy that holds in the 2.3. Subwedge Splitting Signal: Distinguishing asthenosphere beneath ocean basins very likely holds be- Between Slab and Subslab Contributions neath subducting slabs as well. Because Jung et al. [2009] [25] As discussed above, the subwedge splitting signal covered only a limited set experimental conditions and could be due to contributions from anisotropy in the slab because deviatoric stresses in their experiments were very itself, anisotropy in the subslab mantle, or both. While a high, the depth at which any B-type olivine fabric transition contribution to splitting from the downgoing slab cannot be might occur in the mantle remains poorly constrained. ruled out in any given subduction zone, anisotropy in the Therefore, we interpret the observed subslab fast polariza- slab does not appear to represent a significant contribution tion directions as being parallel to the direction of mantle to the global splitting signal, for several reasons. First, the flow; for the vast majority of subduction zones globally, this overwhelming preponderance of trench-parallel fast direc- corresponds to trench-parallel subslab flow. tions in the global data set argues strongly for a mechanism that is related to trench/slab geometry (such as subslab 3. A Synoptic Model for Subduction Zone mantle flow) rather than a mechanism that is related to Anisotropy and Mantle Flow fossil anisotropy in the downgoing lithospheric slab. Second, as we will discuss in section 4, there is no global correlation [27] Our working model for the interaction of subducting between subwedge split times and the age (and thus slabs with mantle flow is described in detail by Long and thickness) of the downgoing slab; this suggests that the Silver [2008]; here we provide a brief summary. Two

8of25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312

experience little or no trench migration, such as the Aleu- tians or Ryukyu, exhibit little or no subslab splitting, while systems that experience significant trench motion, such as Tonga-Kermadec, tend to exhibit strong trench-parallel splitting beneath the slab. This relationship appears to hold regardless of whether the trench is advancing or retreating in the hot spot reference frame. These observations led us to propose a model in which subslab anisotropy is primarily controlled by trench-parallel flow induced by the migration of the trench relative to the surrounding mantle. Such a model had previously been proposed for regions such as South America [Russo and Silver, 1994; Polet et al., 2000]; Figure 3. Histogram of angles between the dominant the global compilation of Long and Silver [2008] provides subslab fast direction and local trench strike for the global evidence that trench-parallel flow beneath slabs, controlled subslab splitting data set. The two trench-perpendicular data by trench migration, is a global phenomenon. points correspond to Cascadia and to the small segment of [28] Our model is summarized in Figures 4 and 5, which the South America subduction zone at a latitude of 20°S. shows a schematic diagram of subslab mantle flow, as well (Preliminary SKS splitting results from Mexico [Stubailo as a plot of subslab split time versus trench migration and Davis, 2007; Le´on Soto et al., 2009] are also consistent velocity similar to that of Long and Silver [2008]. We have with trench-perpendicular subslab splitting, but these results added new data to this plot, taking advantage of measure- are not included in the plot.) The trench-oblique points ments from the Scotia [Mu¨ller et al., 2008], Hikurangi correspond to Japan and to a segment of the Indonesia- [Audoine et al., 2004], Izu-Bonin [Wirth and Long, 2008], Sumatra subduction zone. the Marianas [Wookey et al., 2005], and South America [Rokosky et al., 2006] subduction zones that were not included in our original compilation. For the plot shown striking observations, strengthened by the additional obser- in Figure 5, we have computed a correlation coefficient of vations made in this study, provided the starting point for R = 0.72 between jVtj and dt; this is significant at greater this previously proposed model. First, the global compila- than a 99% confidence level (that is, the range of R values tion of subslab splitting parameters presented here reveals allowed by the data at a 99% confidence level does not that the subslab fast directions are overwhelmingly trench- include zero). parallel (Figures 2 and 3). A few systems, such as Japan and [29] Our model requires trench-parallel flow beneath the South America, exhibit both trench-parallel directions and slab without producing significant slab-entrained flow and directions that are oblique or perpendicular to the trench, but thus entails three significant properties. First, the suppres- even these regions tend to be dominated by trench-parallel sion of slab-entrained flow requires a mechanism for anisotropy. The one notable exception in the published decoupling downgoing slabs from the mantle material literature is Cascadia (this exception is discussed in more beneath it. Possible mechanisms for and implications of detail in sections 6 and 8). Second, there is a systematic this decoupling, along with the exceptions represented by relationship between subslab split time (which is a rough regions such as Cascadia, are explored in section 6. Second, proxy for the strength of anisotropy) and trench migration our model requires a barrier to flow at depth, likely at either velocity in a hot spot reference frame (as compiled by the top or the base of the mantle transition zone, which is Heuret and Lallemand [2005]). Subduction systems that permeable to subducting slabs but does not permit the flow

Figure 4. Sketch of the model explored in this paper for subslab anisotropy controlled by trench migration and slab decoupling [after Long and Silver, 2008].

9of25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312

Figure 5. Subslab split time versus the absolute value of trench migration velocity jVtj. The trench migration velocities in a Pacific hot spot reference frame are obtained from the compilation of Heuret and Lallemand [2005]. In this figure and Figures 6–8, error bars represent the range of values allowed by the data (as described by Long and Silver [2008]), and symbols represent different subduction systems according to the legend. Retreating trenches are shown in blue, and advancing trenches are shown in red. The calculated correlation coefficient is R = 0.72. A similar plot is given by Long and Silver [2008]; our updated compilation, however, contains several additional points. of surrounding mantle material. Because the surrounding average subslab splitting parameters, compiled as described mantle cannot penetrate this barrier, it is forced to flow in section 2, and a wide variety of parameters that describe the horizontally by the motion of the trench rather than under- kinematics, behavior, and morphology of subduction zones neath the slab. The presence of such a barrier to entrained and may be relevant to the subslab flow. Our previous work flow has important implications for global mantle circula- [Long and Silver, 2008] focused on the relationship between tion and mass transfer between the lower and upper mantle; splitting parameters and trench migration velocity. Here, we these are discussed in section 7. Third, in order to force present a more complete assessment of any relationship mantle material to flow parallel to the trench, a distant between subslab shear wave splitting and parameters that barrier to horizontal flow is also required (in the absence of describes subduction. As discussed in section 2, the vast such a barrier, a moving trench would simply force mantle majority of subduction zones worldwide exhibit predomi- to translate along with it in the trench migration direction, nantly trench-parallel fast directions, with one major excep- rather than in a trench-parallel direction). This is demon- tion (Cascadia) and a few that also exhibit regions of oblique strated by numerical and analog box modeling studies; the (Japan, Sumatra) or trench-perpendicular (South America) boundaries of the box are required to induce trench-parallel fast directions. Because the global subslab splitting signal is flow [e.g., Buttles and Olson, 1998; Kincaid and Griffiths, overwhelmingly dominated by trench-parallel fast directions, 2003]. Such a barrier may be provided by the global system we focus on the subslab delay time dt as the splitting of subduction zones. Indeed, in the Pacific, where most parameter that we compare to subduction parameters. We contemporary subduction is taking place, the entire Pacific compare subslab dt ranges to a total of 11 parameters (in Rim of subduction zones likely represents a closed system addition to the trench migration velocity, described in [Husson et al., 2008], acting as distant boundaries that act as section 3) that describe subduction: total convergence rate, barriers to trench-normal motion. normal convergence rate, convergence obliquity angle, age of subducting lithosphere, slab dip, arc curvature, arc 4. Comparison of Splitting Parameters With length, maximum depth of seismicity, maximum depth of Other Tectonic Parameters slab penetration (from tomographic models), descent rate of the slab into the mantle, and the so-called thermal parameter, [30] Comparisons among different subduction parameters described below. In our companion study of global patterns have shown to be useful both in characterizing relationships of mantle wedge anisotropy, currently in progress, we also between sets of parameters and in elucidating the physical examine subduction parameters that may be related to processes that produce those relationships [e.g., Jarrard, processes in the wedge, such as overriding plate strain, 1986; Carlson, 1995; Heuret and Lallemand, 2005; Cruciani overriding plate motion, volcanic production, and the depth et al., 2005; Syracuse and Abers, 2006]. One promising to the slab beneath active volcanoes. approach, therefore, to understanding the processes that [31] Several recent compilations of subduction zone control seismic anisotropy is to undertake comparisons parameters have appeared, including those by Heuret and between shear wave splitting parameters and parameters that Lallemand [2005], Lallemand et al. [2005, 2008], and describe subduction. Here we present comparisons between Syracuse and Abers [2006]. We used values of total

10 of 25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312 convergence rate, normal convergence rate, convergence cates that even when the age of the subducting lithosphere is obliquity, subducting plate age, descent rate, and the thermal corrected for the effect of different path lengths due to slab parameter from Syracuse and Abers [2006]. The thermal dip, there is no statistically significant correlation between parameter f is defined as the product of the plate age and slab age and subwedge delay time. the descent rate of the slab [Kirby et al., 1996; Syracuse and [34] The comparisons between subslab split time and Abers, 2006]. We relied on estimates of arc curvature from subduction zone parameters shown in Figures 5 and 6 Tovish and Schubert [1978]. Our estimates for the maxi- support the Long and Silver [2008] model for trench- mum depth of seismicity in each subduction zone were parallel flow beneath subducting slabs controlled by trench obtained from the slab contours described by Gudmundsson migration. The new subduction zones (Izu-Bonin, Marianas, and Sambridge [1998] and Syracuse and Abers [2006]. For South Sandwich, and New Zealand) that we have added to estimates of the arc length, we used a compilation of ‘‘slab our compilation fit the linear trend in subslab split time with width’’ values from Schellart et al. [2007] (roughly equiv- trench migration velocity well (Figure 5), and the absence of alent, in their definition, to the length of the arc); however, other notable global correlations (Figure 6) between subslab we have separated the Izu-Bonin-Marianas system and the split time and parameters that describe subduction lend Japan-Kurile-Kamchatka system into two separate arcs for support to this model. There are, as we have noted, a few the purpose of calculating arc length. We also examined exceptions to the global trend of trench-parallel fast direc- the maximum depth of slab penetration as compiled by tions, such as the consistent trench-perpendicular fast direc- Lallemand et al. [2005] from various tomographic models; tions observed in Cascadia; these exceptions are discussed although there is, certainly, a great deal of subjectivity in the in the context of our model in sections 6 and 8. We also note interpretation of tomographic images of slabs, many studies that although the global subslab flow field appears to be have distinguished between slabs that are stagnant in the controlled by trench migration, other effects likely modify transition zone and slabs that penetrate into the lower mantle this flow field locally. For example, complex slab morphol- [e.g., Fukao et al., 2001], and we regard this as an important ogy likely has an effect on the 3-D subduction flow field. descriptive parameter for the slab. This has been explored by Kneller and van Keken [2007, [32] Plots of subslab delay times (dt) against each of the 2008], although they restricted their investigation to the subduction parameters described above are presented in mantle wedge region. Locally complex slab morphology Figure 6; the relationship between dt and trench migration likely has a significant effect on local fast directions; for velocity in a hot spot reference frame, as compiled by example, the trench-perpendicular fast directions observed Heuret and Lallemand [2005], is shown in Figure 5 and in a localized region in South America [Polet et al., 2000] was discussed in section 3 and by Long and Silver [2008]. coincide geographically with a change in the arc morphol- As with Figure 5, we have computed the correlation ogy and probably a change in slab morphology as well. coefficient for each relationship shown in Figure 6. For With these caveats, however, the comparisons presented in each of the plots presented in Figure 6, we have not used the Figures 5 and 6 appear to be consistent with our previously split time estimates for a given subduction zone if there is proposed subslab flow hypothesis [Long and Silver, 2008]. not a corresponding estimate for the subduction parameter under study. As Figure 6 demonstrates, we have not 5. Reference Frames for Trench Migration and identified any other striking global correlations between for the Convecting Mantle subslab split time and parameters that describe subduction. None of the correlations tested here is significant at the 95% [35] Our model invokes the migration of trenches confidence level, in contrast to the correlation with jVtj, relative to the ambient surrounding mantle to induce a which is significant at the 99%+ confidence level. The three-dimensional flow field that results in dominantly most striking relationship that emerges from our compar- trench-parallel flow beneath slabs. A correct characteriza- isons, therefore, is that between subslab split time and tion of the relative migration thus requires the specification trench migration velocity in a Pacific hot spot reference of this mantle motion. If the motion is common to all frame. subduction zones, then it would correspond to a single [33] A key question, as discussed in section 2, is whether reference frame. The characteristics of this mantle motion, the subwedge delay time correlates with slab age (a proxy however, remain poorly known. Until now, we have used for slab thickness), which could potentially indicate a trench migration velocities in a Pacific hot spot reference primary contribution from anisotropy in the slab itself. As frame [Gripp and Gordon, 2002] to describe trench motion, shown in Figure 6d, there is no correlation between sub- but the relationship shown in Figure 5 should depend on wedge delay time and the age of the subducting lithosphere. reference frame. The magnitude and, in some cases, the This simple comparison does not take into account varia- direction of trench migration depends heavily on the plate tions in path length through the slab due to differences in motion model and reference frame used [e.g., Stegman et slab dip, however. Assuming nearly vertical SKS propaga- al., 2006; Schellart et al., 2007, 2008; Lallemand et al., tion, the range of slab dips contained in our data set (10°– 2008]. For example, the central part of the South America 65°) would cause a variation in slab path length up to a subduction zone is retreating, with an average velocity of factor of 2.3. In order to more fully investigate this effect, 35 mm yr1, in a Pacific hot spot reference frame, while it we have carried out an additional test (similar to that shown is nearly stationary in an Indo-Atlantic hot spot reference in Figure 6d) in which the age of the slab was ‘‘corrected’’ frame and is advancing slowly (10 mm yr1) in a no-net- by dividing by a path length factor calculated from the slab rotation reference frame [Schellart et al., 2008]. The central dip before testing for a statistical correlation. This test and northern parts of the Ryukyu subduction zone are nearly yielded a correlation coefficient of R = 0.18, which indi- stationary in a Pacific hot spot reference frame, but are

11 of 25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312

Figure 6. Subslab split time plotted against a variety of parameters describing subduction: (a) total convergence velocity, (b) trench-normal convergence velocity, (c) convergence obliquity angle, (d) age of subducting lithosphere, (e) slab dip, (f) arc curvature, (g) arc length, (h) maximum depth of seismicity, (i) maximum depth of slab penetration from tomography, (j) slab descent rate, and (k) thermal parameter. Values for each subduction parameter were compiled from the sources described in the text. Symbols represent different subduction systems according to the legend in Figure 5. The calculated correlation coefficients are R = 0.28 for Figure 6a, R = 0.27 for Figure 6b, R = 0.20 for Figure 6c, R = 0.01 for Figure 6d, R = 0.06 for Figure 6e, R = 0.23 for Figure 6f, R = 0.02 for Figure 6g, R = 0.04 for Figure 6h, R = 0.39 for Figure 6i, R = 0.26 for Figure 6j, and R = 0.10 for Figure 6k.

12 of 25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312

Figure 6. (continued) retreating with velocities of 20 mm yr1 in the Indo- questions: first, does the relationship we identified between Atlantic hot spot and no-net-rotation reference frames subslab dt and jVtj (the absolute value of the trench [Schellart et al., 2008]. It is, therefore, imperative to test migration velocity) in the Heuret and Lallemand [2005] the comparisons between trench migration rate and the compilation hold if we use trench migration estimates for a strength of subslab anisotropy established here in different variety of reference frames and plate motion models? Can reference frames. This will allow us to answer several we identify a different reference frame that improves the fit

13 of 25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312 of the relationship we observe in the Pacific hot spot variety of different plate motion models and reference reference frame? And might this, in principle, allow us to frames (Figure 7). We used the estimates of trench migra- identify the best reference frame for the convecting mantle, tion velocities compiled by Schellart et al. [2008] for eight at least as it relates to subduction zones (which are located different combinations of reference frames and plate motion mainly around the Pacific basin)? models: (1) the plate motion model and Pacific hot spot [36] In order to answer these questions, we have repro- reference frame of Gripp and Gordon [2002], which was duced the plot of dt versus jVtj shown in Figure 5 for a also used in the compilation of Heuret and Lallemand

Figure 7

14 of 25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312

Figure 7. (continued)

[2005] and by Long and Silver [2008] (Figure 7a), (2) the frame proposed by Steinberger et al. [2004] based on an plate motion model and fixed hot spot reference frame of improved reconstruction of the Emperor-Hawaii seamount Gordon and Jurdy [1986] (Figure 7b), (3) the no-net-rotation chain using a mantle convection model. The Steinberger et model of Argus and Gordon [1991] (Figure 7c), (4) the al. [2004] reference frame and plate motion model incor- no-net-rotation model of Kreemer et al. [2003] (Figure 7d), porates an updated hot spot frame that accounts for moving (5) the plate motion model of DeMets et al. [1994] in hot spots. These three comparisons are shown in Figure 8. the Antarctic plate reference frame of Hamilton [2003] [38] The plots of subslab dt versus jVtj (Figures 7 and 8) (Figure 7e), (6) the plate motion model of DeMets et al. demonstrate that the Pacific hot spot based reference frames [1994] with the Wessel et al. [2006] Pacific hot spot described by Gripp and Gordon [2002], Wessel et al. reference frame (Figure 7f), (7) the plate motion model of [2006], and Steinberger et al. [2004] yield the strongest DeMets et al. [1994] in the Indo-Atlantic hot spot reference correlations between subslab anisotropy and trench migra- frame of O’Neill et al. [2005] (Figure 7g), and (8) the plate tion. The correlations between subslab dt and jVtj for the motion model of Kreemer et al. [2003] in the O’Neill et al. Gripp and Gordon [2002] reference frame are significant at [2005] Indo-Atlantic hot spot reference frame (Figure 7h). the 99% confidence level (using both the Heuret and Essentially, the differences among these eight sets of trench Lallemand [2005] and the Lallemand et al. [2008] compi- migration velocity values can be attributed either to differ- lations) and the 93% confidence level (using the Schellart et ences in the reference frame, or differences in the model for al. [2008] compilation), while the correlation for the Wessel back-arc deformation. Further details on this compilation et al. [2006] reference frame (Figure 7f) is significant at the are given by Schellart et al. [2008]. 91% confidence level and that for the Steinberger et al. [37] We have also reproduced our dt versus jVtj compar- [2004] reference frame is at 88%. All of the correlations that isons using the three different compilations of trench are significant above the 88% confidence level, therefore, migration velocities given by Lallemand et al. [2008]: the are associated with Pacific hot spot reference frames. All of HS3 Pacific hot spot reference frame [Gripp and Gordon, the other plate motion models and reference frames (no-net- 2002] used by Heuret and Lallemand [2005], the no-net- rotation, Indo-Atlantic hot spot, and Antarctic) yield no rotation frame of DeMets et al. [1994], and the reference

Figure 7. Subslab split time versus absolute value of trench migration velocity for each of the plate motion model and reference frame combinations described by Schellart et al. [2008] and in section 5. (a) Plate motion model and Pacific hot spot reference frame of Gripp and Gordon [2002]. (b) Plate motion model and hot spot reference frame of Gordon and Jurdy [1986]. (c) The no-net-rotation reference frame and plate motion model of Argus and Gordon [1991]. (d) The no-net- rotation reference frame and plate motion model of Kreemer et al. [2003]. (e) The plate motion model of DeMets et al. [1994] in the Antarctic plate reference frame of Hamilton [2003] (f) The plate motion model of DeMets et al. [1994] in the Wessel et al. [2006] Pacific hot spot reference frame. (g) The plate motion model of DeMets et al. [1994] in the Indo- Atlantic hot spot reference frame of O’Neill et al. [2005]. (h) The plate motion model of Kreemer et al. [2003] in the O’Neill et al. [2005] reference frame. Symbols represent different subduction systems according to the legend in Figure 5. The calculated correlation coefficients are R = 0.40 for Figure 7a, R = 0.24 for Figure 7b, R = 0.19 for Figure 7c, R = 0.06 for Figure 7d, R = 0.28 for Figure 7e, R = 0.38 for Figure 7f, R = 0.23 for Figure 7g, and R = 0.04 for Figure 7h. The splitting estimate from GSN station PSI in the Tonga-Kermadec subduction zone is shown with a dashed line; accurate trench migration velocity estimates are difficult to make for this station because of the very large along-strike gradients in trench migration (the station is located near the transition from rapid trench retreat to the north to rapid trench advance to the south).

15 of 25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312

Figure 8. Subslab split time versus absolute value of trench migration velocity for each of the reference frames described in Lallemand et al. [2008]. (a) The Pacific hot spot reference frame of Gripp and Gordon [2002]. (b) The reference frame of Steinberger et al. [2004]. (c) The no-net-rotation reference frame [DeMets et al., 1994; Gripp and Gordon, 2002]. Symbols represent different subduction systems according to the legend in Figure 5. The calculated correlation coefficients are R = 0.66 for Figure 8a, R = 0.36 for Figure 8b, and R = 0.07 for Figure 8c. As in Figure 7, the splitting estimate from GSN station PSI in the Tonga-Kermadec subduction zone is shown with a dashed line.

discernable relationship between dt and jVtj. If our model [39] In principle, the compilations of Lallemand et al. for subslab flow is correct, the fact that the Pacific hot spot [2008] and Schellart et al. [2008] for the Gripp and Gordon reference frames yield the clearest linear relationship [2002] Pacific hot spot reference frame should be similar, between subslab dt and jVtj suggests that such reference although in practice different estimates for back-arc defor- frames do the best job of describing a ‘‘rest’’ frame for mation can result in incompatible estimates for jVtj.In Earth’s subduction zones relative to the upper mantle. This particular, we note that the correlation between subslab dt is perhaps unsurprising, as nearly all subduction zones are and jVtj in the Pacific hot spot reference frame [Gripp and located either on the margins of the Pacific plate (e.g., Gordon, 2002] is poorer using the Schellart et al. [2008] Aleutians, Kuriles, Japan, Izu-Bonin-Marianas, Tonga) or in compilation than that of Lallemand et al. [2008] (Figure 7a geographical proximity to it (e.g., Central America, Carib- versus Figure 8a), with a correlation coefficient of R = 0.66 bean, Scotia, Ryukyu, Indonesia). The sole exception is the for Lallemand et al. [2008] versus R = 0.40 for Schellart et Calabria subduction zone in southern Italy. Because there is al. [2008]. This is mainly due to the very different trench only one truly ‘‘non-Pacific’’ subduction zone in the com- migration velocities obtained at the location of GSN station pilation, it is difficult to evaluate whether the correlations PSI in the Kermadec subduction zone, at which splitting with Pacific hot spot trench migration velocities would was measured by Long and Silver [2008], although a few improve if subduction zones located far from the Pacific other subduction systems (Izu-Bonin, Java-Sumatra, and plate were removed from the plots shown in Figures 7 and 8. Kamchatka) also yield dissimilar estimates using the different compilations. The disparity obtained for the

16 of 25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312

Kermadec subduction zone is mainly due to the unique it is possible that these two results reflect real hemispherical kinematics of the Tonga-Kermadec subduction system; the variability in flow in the upper mantle. Our own criterion for northern part of the subduction zone is retreating rapidly in the ‘‘best’’ reference frame is heavily biased to the Pacific most reference frames, while the southern part is advancing; basin, since the subduction zones in our compilation are station PSI is located near the transition from trench retreat nearly all located on or near the margins of the Pacific to trench advance. The different back-arc deformation Ocean. In contrast, the approach taken by Becker [2008] estimates and along-strike spatial sampling used by Schellart likely effectively downweights the Pacific. Because the et al. [2008] versus Lallemand et al. [2008] results in Pacific is a fast moving plate, the orientation of astheno- disparate estimates for jVtj at this station location. Physically, spheric shear (representing the difference between plate however, one would expect that the rapid trench movement to velocity and subasthenospheric mantle velocity) and there- the north and south of this ‘‘hinge’’ point would result in fore seismic anisotropy, is insensitive to the motion of the considerable trench-parallel flow; taking this into account underlying subasthenospheric mantle [Behn et al., 2004]. would improve the fit shown in Figure 7a. For slow moving plates, such as the Atlantic, the anisotropy [40] In general, dissimilar back-arc deformation estimates is, in contrast, much more sensitive to the subasthenospheric (which variously incorporate geological and geodetic data), velocity, and these regions would consequently dominate treatments of tectonic erosion/accretion (which is ignored in any global fit. the Lallemand et al. [2008] compilation), choices about [43] We therefore hypothesize that these two studies, which microplates and/or arc blocks to consider, and along- taken together, represent actual hemisphere-scale variability strike sampling variations between the two compilations can in upper mantle flow. This may be linked to structures in the account for the differences between the two compilations, lowermost mantle; the existence of prominent, large-scale many of which are subtle (see Schellart et al. [2008] for a low-velocity anomalies beneath the Pacific (the ‘‘Pacific more complete discussion of the differences among trench Anomaly’’) and beneath Africa (the ‘‘African Anomaly’’) migration velocity compilations). For example, Schellart et has been firmly established [e.g., Garnero and McNamara, al. [2008] relied on geological estimates of back-arc defor- 2008; Ni et al., 2002; Wang and Wen, 2007]. Such structures mation rates wherever possible in their Pacific hot spot may have a dramatic effect on hemisphere-scale mantle reference frame compilations, while Lallemand et al. [2008] convection; such an effect has been studied by, e.g., relied mainly on geodetic estimates. Despite the (often Gaboret et al. [2003]. If a Pacific hot spot reference frame subtle) differences between jVtj estimates (with the excep- is, indeed, the most appropriate ‘‘rest’’ frame for Pacific tion of Tonga-Kermadec, all of the jVtj estimates for mantle flow, and either the no-net-rotation or the Indo- individual subduction zones shown in Figures 7a and 8a Atlantic hot spot reference frame is most appropriate for the differ by less than 10–15 mm yr1), statistically signif- Atlantic-African hemisphere, this implies a relative motion icant correlations are observed for the Gripp and Gordon between the two hemispheres (the net rotation of the [2002] reference frame for both compilations that we lithosphere in the HS3-Nuvel-1A model is 0.436° Myr1 examined. [Gripp and Gordon, 2002; Becker, 2008]). Our model does [41] The dissimilarities in local trench migration veloci- not constrain a mechanism for this hypothesized relative ties between different compilations and the unique kinemat- motion, but our suggestion seems to be consistent with ics of the Tonga-Kermadec subduction system both bring up previous studies that have suggested that Pacific hemisphere a larger point about the utility of localized trench migration mantle convection is largely driven by a hemisphere-scale velocity estimates as an indicator of subslab mantle flow. large upwelling anchored beneath the South Pacific super- Particularly for subduction systems for which the local swell [Gaboret et al., 2003] or that large-scale lower mantle trench migration velocity changes rapidly along strike, such return flow has driven net motion between the Hawaii- as Tonga, the local value of jVtj may not actually be the best Emperor hot spot and the Indo-Atlantic hot spots [e.g., indicator of subslab flow–rapid trench retreat or advance on DiVenere and Kent, 1999; Tarduno et al., 2003, 2009]. The adjacent sections of the trench may induce considerable need to invoke different reference frames for the Pacific trench-parallel flow below trench segments that are them- Ocean basin and the Indo-Atlantic oceans in order to selves nearly stationary. Future modeling work should help explain observations of seismic anisotropy beneath ocean to answer questions about how along-strike gradients in jVtj basins and beneath subducting slabs is, in fact, similar to the affect the subslab flow field and incorporating information need to invoke different reference frames for Pacific versus about such along-strike gradients into plots such as those Indo-Atlantic hot spots. shown in Figures 7 and 8 may help to reconcile discrep- [44] Another approach to determining a mantle rest frame ancies between different trench migration compilations. has been provided by Schellart et al. [2008], who estab- [42] There have been other attempts to define a ‘‘best’’ lished constraints based on a set of assumptions about reference frame for mantle flow, and it is constructive to trench migration behavior. In particular, they sought to compare our results to them. The study closest to our own is minimize the amount of trench migration in the center of that of Becker [2008], who compared the global constraints wide subduction zones, to maximize trench migration at on azimuthal anisotropy in the upper mantle from surface subduction zone edges, and to maximize the number of wave models to the asthenospheric anisotropy predicted by trench segments that retreat (as opposed to advance). numerical models of mantle flow. He concluded that the fit Schellart et al. [2008] concluded that the Indo-Atlantic between predicted and observed asthenospheric anisotropy hot spot reference frame best fits this set of assumptions. was best for a no-net-rotation (or minimal-net-rotation) The approach taken by Schellart et al. [2008] differs reference frame. While this result appears to be at odds markedly from that taken in this study, as ours relies on with our preference for the Pacific hot spot reference frame, measurements of the subslab mantle flow field (which we

17 of 25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312

Figure 9. Cartoons of two scenarios for the fate of the asthenosphere at subduction zones. (a) An asthenosphere that is controlled solely by pressure (depth); in this scenario, the low viscosity of the asthenosphere results from the proximity of the adiabatic temperature to the mantle solidus. If the depth extent of the asthenosphere is controlled only by pressure, then the asthenosphere will ‘‘pinch out’’ beneath the subducting slab and there is no mechanism for subslab decoupling at subasthenospheric depths. (b) A hot, buoyant asthenosphere that is formed by shear heating; a thin layer of such an asthenosphere is entrained beneath the slab, in a scenario similar to that envisioned by Phipps Morgan et al. [2007]. interpret as being due to trench migration) rather than combination with a significant trench-parallel component assumptions about trench migration behavior. of deformation seems physically unworkable. [47] Another possibility is that there is a component of 6. Dynamics of the Subslab Mantle During entrained flow beneath the slab, but the strain associated Subduction: Models and Implications with this entrained flow does not lead to observable anisot- ropy and is thus ‘‘hidden.’’ This would be possible if strain [45] The observation that globally, subslab fast directions in this region were accommodated by diffusion creep, which are nearly always parallel to the trench appears contrary to does not produce lattice preferred orientation (LPO) and the expectations of a simple, 2-D model of entrained flow therefore seismic anisotropy, instead of diffusion creep, beneath downgoing slabs. This lack of evidence for which does produce LPO [e.g., Karato and Wu, 1993]. entrained flow beneath slabs (with a few exceptions, such Again, however, constraints are provided by the observation as Cascadia) has led us to consider a variety of models that of significant trench-parallel anisotropy in most subduction could result in dominantly trench-parallel subslab flow zones globally. A scenario in which entrained flow directly directions. A closely related issue is the fate of suboceanic beneath the slab is accommodated by diffusion creep would asthenosphere as a subducting plate begins its descent into have to accommodate a region of trench-parallel flow in the the mantle, as we will show. Shear wave splitting behavior dislocation creep regime below the entrained flow layer, and beneath oceanic plates far away from trenches or ridges is a mechanism for a transition from diffusion creep to relatively well understood as being due to simple shear in dislocation creep with increasing depth is difficult to envi- the asthenosphere, with the geometry of the fast directions sion. A more plausible ‘‘hidden’’ layer of entrained flow is being controlled by the vectorial difference between sub- one that is thin enough (on the order of 10 km) so that any asthenospheric motion and plate motion [e.g., Behn et al., anisotropy is invisible to shear waves passing through it; as 2004; Conrad et al., 2007]. It is not at all clear, however, explained in more detail below, this is the scenario we favor. what the fate of the asthenosphere is at subduction zones, We note that either case (the ‘‘thick’’ layer of entrained and this is a major consideration in the models we consider asthenospheric flow accommodated by diffusion creep, or to explain the apparent lack of entrained flow beneath slabs. the ‘‘thin’’ layer of entrained asthenosphere) requires the [46] First, we consider if there is any way to reconcile the advection of asthenospheric material that has distinct prop- preponderance of trench-parallel fast directions with an erties from the ambient mantle. If the asthenosphere merely entrained flow model. One possibility would be a model represents the depth range over which the temperature is that invokes perfectly entrained flow, that is, in which sufficiently close to the mantle solidus to cause significant downgoing slabs do entrain material, but there is no strain weakening (that is, a reduction in viscosity), then it would involved. In this scenario, there is a translation of mantle be impossible for slabs to advect asthenospheric material material down with the slab, along with a small component deeper into the mantle; the physical properties of such an of trench-parallel strain that gives rise to the observed asthenosphere would be controlled only by depth (pressure) trench-parallel fast directions. However, a perfect transla- (Figure 9). tion of mantle material beneath the slab, that is, entrainment [48] Our preferred model for the subslab mantle is that without deformation, would imply stress-free boundary downgoing slabs generally entrain a thin layer of astheno- conditions, and the fact that there is significant anisotropy sphere that acts as a decoupling zone between the slab and (and therefore deformation) in many subslab regions the subslab mantle, which allows for the motion of the contradicts this implication. Therefore, a model that trench/slab system to induce trench-parallel flow (Figure 9). invokes strain-free entrainment of mantle material in The viability of this model depends on the physical (and

18 of 25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312 perhaps chemical) characteristics of the asthenosphere, but in terms of a dimensionless temperature perturbation q = 0 it requires an asthenosphere that is physically distinct from goT /To(go = Ea/RTo) and ~x = x/h. Br is the Brinkman the surrounding ambient mantle. One possible mechanism is number, defined as [Turcotte and Schubert, 1982] that suggested by Phipps Morgan et al. [2007]; they 2 2 conducted numerical and laboratory experiments and t h go showed that a thin (10–30 km) layer of hot, buoyant Br ¼ ; ð3Þ kmoTo asthenosphere may be entrained beneath downgoing slabs. This interpretation depends, however, on a particular view where mo is the reference viscosity at temperature To. The of the asthenosphere [Phipps Morgan et al., 1995]: namely, solution to (2) yields the temperature perturbation as a that is a consequence of positively buoyant upwelling function of depth. This solution can be used to relate Br to plumes in the mantle that feed hot, chemically depleted the maximum temperature perturbation in the layer, q(1) material into the sublithospheric mantle. In this scenario, the [see Turcotte and Schubert, 1982, Figure 7–17]. This asthenosphere is intrinsically buoyant, has a higher potential temperature, however, is a double-valued function of Br, temperature and is much weaker than the mantle beneath it. defining a high-temperature and low-temperature branch. [49] The entrainment of a thin layer of buoyant astheno- The correct temperature can be calculated by specifying the sphere beneath slabs does not, however, necessarily require velocity of the upper plate, u. Since u is the integral of the a plume-fed, chemically depleted asthenosphere; here, we strain rate, e_, through the layer, i.e., propose an alternate view of the asthenosphere that is consistent with such an entrainment scenario. For a model Zh Z1 of the type presented by Phipps Morgan et al. [2007] invoking a thin entrained subslab asthenospheric layer to u ¼ e_ðÞx dx ¼ h e_ðÞ~x d~x; be viable, the asthenosphere must be buoyant. One possible 0 0 mechanism for this is shear heating. We envision a scenario in which mantle material beneath newly created oceanic we can use the relation for a Newtonian fluid, t = me_,to 1 1 q lithosphere, subjected to high-strain shear deformation in- write e_ = tm = tmo e so that duced by the motion of the overlying oceanic plate, under- goes shear heating as it moves progressively away from the Z1 1 q 1 ridge. Numerical models of corner flow in the vicinity of a u ¼ thmo e d~x thmo m: spreading ridge [e.g., Nippress et al., 2007] indicate that the 0 sublithospheric mantle is subjected to strains much larger than the mantle material directly beneath the ridge. If the (Here m denotes the integral expression in the previous viscosity of the starting material is large enough, the heating 2 equation.) Substituting t = mou/hm in (3) yields Br = mou go/ (and associated lowering of the viscosity) that would result 2 2 2 km To. Writing as Brm = mou go/kTo, the right-hand side from viscous dissipation in the sublithospheric mantle may consists of nominal parameters, and one can then determine be nonnegligible. the unique point on the Br q(1) curve that satisfies this [50] The approximate effects of shear heating produced equation, which then defines the maximum temperature rise. by asthenospheric flow can be estimated by obtaining the [51] A reasonable set of values for the upper mantle (a steady state temperature profile for Couette flow with both temperature of 1500 K at the base of the lithosphere, an viscous dissipation and temperature-dependent viscosity. initial (reference) viscosity of 1021 Pa s, a plate velocity of Here we briefly describe the equations derived by Turcotte 1 1 1 60 mm yr , go = 20, and k =4Wm K ) places us on and Schubert [1982] and discuss our estimate of the size of the high-temperature branch, and we would predict a the shear heating effect using nominal values for mantle temperature increase of over 100 K and a viscosity decrease viscosity and temperature. Following Turcotte and Schubert of 1 order of magnitude. (For a more detailed derivation [1982, pp. 320–324], the temperature, T, resulting from of the relationship between the Brinkman number and the Couette flow in a channel of thickness h driven by a stress t steady state temperature profile in the channel, we refer the applied to the upper boundary and with fixed lower bound- reader to Turcotte and Schubert [1982]. We are also ary can be described by currently undertaking further analytical and numerical mod- eling of asthenospheric shear heating.) In this model, d2T t2 therefore, newly formed suboceanic asthenosphere (after k þ ¼ 0 ð1Þ dx2 m steady state is reached) would have a viscosity of about 1019 – 20 Pa s (depending on the values used in the calcula- tion), consistent with estimates from geophysical observa- where m is the temperature-dependent viscosity of the form tions such as postglacial rebound [e.g., Fjeldskaar, 1994] or m = CeEa/RT, E is the activation energy, R is the ideal gas a triggered earthquakes [e.g., Rydelek and Sacks, 1988]. The constant, and C is a constant. Boundary conditions are thermal buoyancy provided by the 100 K temperature rise imposed such that T = T on the upper boundary and 0 would make the entrainment of a thin decoupling layer dT/dx = 0 at the lower boundary (insulating). Assuming that beneath the slab feasible. the temperature perturbation, T 0, due to shear heating is [52] Our shear heating model for the origin of the small compared to T , (1) can be rewritten as o asthenosphere makes testable predictions that should pro- vide additional means for discriminating among different d2q models for the mechanism which gives rise to subslab þ Breq ¼ 0 ð2Þ d~x2 decoupling. Our model implies a starting viscosity value

19 of 25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312 for protoasthenosphere that is the same as that of ambient [54] The suggestion that the asthenosphere beneath oce- mantle material; the reduction in viscosity to 1019 – 20 Pa s anic plates has characteristics (such as excess temperature or is accomplished by shear heating through progressive strain partial melt) that allow it to be entrained beneath subducting that results in a temperature rise of 100 K. A prediction of slabs is consistent with recent studies of the lithosphere- our model, therefore, is that the asthenosphere will have an asthenosphere boundary (LAB) beneath oceans [Rychert excess potential temperature (that is, the temperature rela- and Shearer, 2009; Kawakatsu et al., 2009]. In particular, tive to an adiabatic temperature profile) of 100 K, and a Kawakatsu et al. [2009] imaged a sharp LAB beneath the key test of our model is whether this prediction can be subducting Pacific slab beneath Japan using receiver func- reconciled with the one-dimensional seismic velocity pro- tions; their preferred model to explain their observations files inferred for the oceanic mantle. Another test can be invokes partial melt as a cause for the sharp reduction in provided by examining lateral variations in seismic veloc- shear wave velocity across the LAB. The proposed exis- ities at asthenospheric depths: our model predicts that tence of partial melt in the asthenosphere is consistent with ‘‘young’’ asthenosphere near spreading ridges that has not the excess temperatures predicted by our shear heating yet undergone sufficient strain to experience shear heating model, and an asthenosphere that contains partial melt should be relatively cooler than fully developed astheno- would be able to be entrained beneath subducting slabs to sphere away from spreading ridges. A key question is provide subslab decoupling. In general, models for subslab whether this is borne out by three-dimensional seismic asthenospheric entrainment, including the shear heating velocity models of the mantle; however, it is questionable model proposed here, raise questions about the fate of the whether global models provide the resolution needed to suboceanic asthenosphere at subduction zones, and, more examine such small-scale lateral variations, particularly broadly, about the nature of the asthenosphere itself. Many since seismic velocities in the immediate vicinity of ridges of the questions raised by the inference of a thin decoupling are strongly affected by the presence of partial melt. For zone beneath subduction zones relate to larger questions example, global surface wave models such as that of Nettles about how the asthenosphere is defined, what controls its and Dziewonski [2008] show low velocities in the immedi- physical properties, and what is its fate when the overlying ate vicinity of spreading ridges and relatively higher veloc- lithosphere begins to descend into the mantle at subduction ities away from the ridges. This is the opposite of the effect zones. Many of these issues are poorly understood, and predicted by the shear heating model; however, the low represent important avenues for further research. velocities beneath ridges are very likely due to the effect of partial melt beneath the ridge. The poor lateral resolution 7. A Constraint on Mass Transfer Between the afforded by such global studies does not allow for the Upper and Lower Mantle? separation of this partial melt effect from the temperature effect predicted by the shear heating model. The paucity of [55] The model proposed here invokes a decoupling zone seismic stations in the oceans means that there are few, if beneath the slab and a partial barrier to flow at depth, as first any, previous regional seismic studies that have both the proposed by Russo and Silver [1994] for South America. resolution and the lateral extent to look for lateral temper- Although there is ample evidence from global seismic ature variations in the vicinity of ridges, but this could tomography that some slabs penetrate the transition zone represent an important scientific target for future studies. and sink into the lower mantle [e.g., van der Hilst et al., [53] While our hypothesized shear heating mechanism 1997; Li et al., 2008], the extent to which such slabs entrain provides one possibility for asthenospheric entrainment, the surrounding mantle, and therefore the amount of mate- there are certainly other possibilities, depending on the rial flux across the transition zone and the degree of mixing physical state of the asthenosphere. For example, it has between the upper and lower mantle, remain hotly debated been suggested that the asthenosphere possesses a high [e.g., Tackley, 2008]. Our model for subslab flow in sub- volatile content [Karato and Jung, 1998] or, alternatively, duction zones suggests that in cases where the downgoing that it corresponds to a minimum in water solubility in slab is decoupled from the underlying mantle, this partial aluminous orthopyroxene [Mierdel et al., 2007]. Other barrier to flow forces upper mantle material to flow parallel possible mechanisms for creating a thin decoupling zone to the trench if the trench is moving in the reference frame beneath downgoing slabs may be related to the power law of the convecting mantle. rheology of the mantle, the possible presence of partial melt [56] If our model correctly describes the subslab flow in the asthenosphere, or mechanical anisotropy, which has field then there are important implications for large-scale been shown to have a significant effect on mantle flow in mantle dynamics, in particular for the amount of mass numerical models [e.g., Lev and Hager, 2008]. We empha- transfer across the transition zone and the extent to which size that the basic requirement of our subslab mantle flow mantle convection is well mixed. The qualitative nature of model is that a thin decoupling zone must be present; we the convecting mantle continues to be vigorously debated; remain open to alternative models that could accomplish end-member models invoke either whole mantle convection this. The class of decoupling models that invoke the and mixing or a nearly impenetrable barrier to flow at the entrainment of a thin layer of weak asthenosphere beneath base of the transition zone that separates the mantle into two the slab includes the shear heating model explored here, but distinct layers. Attempts to reconcile geochemical and could also include models in which the asthenosphere is geophysical observations have resulted in a range of inter- defined by the presence of partial melt or is rich in volatiles. mediate models, many of which involve exchange between The shear heating model is one possibility, but it is not the the upper and lower mantle which is limited to slab flux only one. only [Silver et al., 1988] or invoke partial or intermittent layering [e.g., Tackley, 2008].

20 of 25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312

[57] Constraints from seismic tomography have been these represent the most extreme examples of the subduc- incorporated into the mantle mixing debate, as the predictions tion of young lithosphere on the present-day Earth. In this made by mantle mixing models are compared to the ‘‘snap- paper, we have proposed a mechanism for the decoupling of shot’’ of present-day mantle structure provided by tomo- slabs and subslab mantle material that invokes shear heating graphic inversions for seismic wave speed [e.g., Tanimoto in the highly deformed asthenosphere and subsequent and Lay, 2000]. However, while global tomographic techni- entrainment of a thin layer of asthenosphere beneath the ques can image structures such as subducting slabs at depth, downgoing slab that serves as a weak, rapidly deforming, they cannot directly image active mantle flow. The con- decoupling zone. This mechanism depends on the astheno- straints on mantle flow provided by observations of seismic sphere experiencing a large amount of shear strain; for the anisotropy are more direct than those provided by tomogra- case of very young seafloor, we hypothesize that the amount phy, and the constraints on the subslab mantle flow field of strain needed for the shear heating mechanism to reach provided by shear wave splitting observations imply that steady state has not yet been reached, and that decoupling even though slabs may penetrate the lower mantle, they are between the slab and the subslab mantle is not accom- unlikely to entrain much surrounding mantle material with plished. them. This, in turn, suggests that the material flux associated [59] If this hypothesis is correct, it has several important with subducting slabs that penetrate into the lower mantle implications. This model makes testable predictions about may simply represent the slab flux. The constraints on the the character of subslab shear wave splitting in regions character of mantle downwellings provided by the observa- where the subducting lithosphere is young, and would also tion that the subslab flow field is dominated by horizontal predict an along-strike transition from trench-perpendicular rather than entrained flow may help to constrain models of to trench-parallel subslab splitting in subduction zones mantle convection and mixing; the apparent restriction on where the subducting lithosphere is generally young and mass transfer between the upper and lower mantle reservoirs there is an along-strike transition in its age. Future studies of suggested by our model could help to discriminate, for along-strike variations in subslab splitting in such systems example, among numerical models of mantle convection provides a direct test of this hypothesis, which should [e.g., Tackley et al., 2005]. The barrier to entrained flow provide further constraints on the viability of our proposed required by our model tends to support mantle convection model for subslab decoupling. Particularly interesting test models that have limited material flux across the 660-km cases could be provided by, for example, the subduction of discontinuity, such as models that invoke ‘‘leaky layering’’ the Cocos Plate beneath southern Mexico and Central (see the overview by Tackley [2008]). Similar support for the America. In this subduction system, the age of the down- idea that mass flux across the 660-km discontinuity is limited going lithosphere varies from just over 5 Myr to the has been offered by studies of the heterogeneity spectrum of northwest to over 20 Ma for the region beneath Guatemala, tomographic models, which suggest that there may be an El Salvador, and northwestern Nicaragua; seafloor ages then abrupt change in this spectrum across the discontinuity at the decrease again to the southeast, with ages of 10 Ma base of the transition zone [e.g., Dziewonski, 2005]. beneath southern Costa Rica. Splitting results from the TUCAN experiment [Abt et al., 2005, 2009] indicate that 8. Discussion subslab anisotropy is trench-parallel beneath the relatively older lithosphere that underlies stations located in Nicar- [58] The observation that shear wave splitting due to agua. Our hypothesis would predict that the subslab split- subslab anisotropy nearly always exhibits fast polarization ting signal would rotate to trench-perpendicular fast directions that are parallel to the trench (Figure 3) is one of directions for the youngest lithosphere in the northwestern- the most robust features of the global splitting data set in most and southeasternmost parts of the Central American subduction zone regions. However, there are several impor- subduction zone. This prediction could be tested by collect- tant exceptions to this rule, and the significant nearly trench- ing and analyzing data from stations located above the perpendicular splitting attributed to subslab flow beneath youngest parts of the subducting Cocos slab. Preliminary the downgoing Juan de Fuca slab in the Cascadia subduc- data from the Rivera subduction zone (where the young tion zone is the most notable and best constrained. Prelim- Rivera microplate, a fragment of the Cocos plate, is subduct- inary observations of strong, trench-perpendicular SKS ing beneath North America) indicate that subslab anisotropy splitting at stations overlying the Middle America subduc- beneath the young (<10 Ma) subducting plate is dominantly tion zone beneath Mexico [Stubailo and Davis, 2007; Le´on trench-perpendicular. If there is, indeed, a transition in Soto et al., 2009], to the north of the region of the Middle subslab fast directions related to the age of the subducting America subduction zone that is included in our compila- lithosphere, then this would place constraints on the ‘‘critical tion, may also provide an example of trench-perpendicular distance’’ from the ridge needed for significant shear heating fast directions beneath a subducting slab. An important of young asthenosphere that would produce subslab decou- question, then, is whether our model for subslab mantle pling. If our model for subslab decoupling is correct, this has flow can accommodate these exceptions. One way that significant implications for our understanding of the nature subduction-parallel entrained flow beneath a subducting and dynamics of the asthenosphere, which are currently slab can be reconciled with our model is if the mechanism poorly understood [e.g., Rychert et al., 2005]. for decoupling the slab from the mantle material beneath it [60] Another fruitful area for future work on the nature is not active. Both Cascadia and the northern and southern and implications of the subslab flow field is the integration ends of the Middle America trench feature the subduction of of shear wave splitting constraints with numerical and very young lithosphere; the lithosphere entering both of analog models of subduction zone flow in the presence of these subduction zones is between 5 and 10 Myr in age, and trench migration and subslab decoupling. The model for

21 of 25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312 subslab mantle flow that we propose here is consistent with the slab due to hydrated faults must be strong enough not the global shear wave splitting data set, but it needs to be only to explain the large subwedge splitting observed in validated using geodynamical models to verify whether it regions such as Tonga-Kermadec, Calabria, or Middle is dynamically plausible and whether the relationships America, but also to counteract (through destructive inter- between trench migration and strength of subslab anisotropy ference) the effect of trench-perpendicular splitting due to can be reproduced. Laboratory models have been successful entrained flow beneath the slab. Finally, we point out that in producing trench-parallel flow above subducting slabs for observations of azimuthal anisotropy beneath oceanic plates systems with retreating trenches [e.g., Kincaid and Griffiths, from both splitting observations and surface wave studies 2003; Funiciello et al., 2003] but such modeling studies are very well explained by active flow in the upper mantle have not emphasized the behavior of the subslab flow field. (that is, simple shear between the oceanic lithosphere and Issues such as the character of the flow field for models the underlying mantle) and that there is no obvious corre- incorporating trench advance, as opposed to trench retreat, lation with the geometry of (presumably hydrated) faults whether laboratory models can reproduce the splitting [e.g., Behn et al., 2004; Conrad et al., 2007; Becker, 2008]. observations, and the effect of a subslab decoupling zone This observation seems to be contrary to the predictions of a on mantle flow fields remain to be addressed. Another key hydrated fault model: if hydrated faults in downgoing slabs question is to what extent trench-parallel strain induced by indeed control the SKS splitting observed above subduction trench migration is localized directly beneath the slab, as zones, then the SKS splitting signal within oceanic plates opposed to being distributed over a larger region of the should presumably be controlled by faults as well. While mantle. Some previous laboratory modeling work [Buttles hydrated faults in the upper 10 km of subducting slabs may and Olson, 1998] has indicated that mantle flow beneath make some contribution to shear wave splitting in some slabs is strong enough to locally produce trench-parallel regions, this mechanism does not appear to be the primary anisotropy, but the character of this subslab flow field will control on SKS splitting patterns in subduction zones depend on the viscosity structure of the mantle, the nature of globally. the far-field boundary conditions, and other variables, and [62] On the basis of only the compilation of subwedge remains to be more thoroughly explored through laboratory anisotropy parameters presented in this paper, we cannot and numerical modeling studies. entirely rule out alternatives to the model we propose such [61] As discussed in section 2, we believe that the as the hydrated fault model of Faccenda et al. [2008] or the subwedge splitting observations are most consistent with model proposed by Jung et al. [2009], in which B-type trench-parallel flow in the subslab mantle, rather than a olivine fabric predominates beneath the slab and the observed primary contribution from the slab itself. However, alterna- trench-parallel fast directions are due to 2-D entrained flow. tive models that invoke anisotropy within the slab have However, our trench-migration-controlled model seems to been proposed; in particular, recent work by Faccenda et al. be the most consistent with the global subwedge splitting [2008] and Healy et al. [2009] has explored the feasibility patterns documented here. Specifically, our model is able to of explaining SKS splitting data with a model for slab explain the variations in subwedge delay time, which varies anisotropy that invokes the hydration and serpentinization globally from 0sto1.5 s or more; the correlation we of faults that penetrate the lithospheric slab. From an have documented between subwedge delay time and trench observational point of view, it is difficult to discriminate migration velocities in a Pacific hot spot reference frame between slab versus subslab anisotropy in the subwedge (Figures 5 and 8) is significant at the 99%+ confidence splitting signal, so most of the arguments about which level. The model proposed by Long and Silver [2008] and model better fits the observations are indirect. For several elaborated upon here can also explain the ‘‘exceptions’’ in reasons, we believe that subslab flow does a better job of the global data set where the subwedge splitting is appar- explaining the global subwedge splitting signal than slab ently trench-perpendicular or oblique, such as Cascadia or anisotropy due to hydrated faults. As discussed in section 4, Mexico. the lack of a correlation between the strength of subwedge splitting and slab age (which correlates with both slab 9. Conclusions thickness and, perhaps, with the extent of fault hydration) argues against a primary contribution to the global sub- [63] We have presented a global compilation of subslab wedge splitting signal from the slab. It is also difficult to splitting parameters and have tested for relationships be- explain the global variability in subwedge split times with a tween those parameters and parameters that describe sub- slab anisotropy model; the strength of subwedge splitting duction, with the aim of determining the dominant varies from little to none (e.g., Ryukyu or the Aleutians) to properties of the subslab flow field on a global scale. We up to 1.5–2 s (Tonga-Kermadec or Calabria), and there is have found that globally, the subslab region is nearly always no obvious mechanism to explain this variability with the dominated by trench-parallel fast directions, and we have hydrated fault model. While Faccenda et al. [2008] argue identified a striking correlation between the strength of that trench-parallel flow beneath subducting slabs is incon- subslab anisotropy and the trench migration rate in a Pacific sistent with the results of analog and numerical modeling hot spot reference frame. We did not identify any other studies, we point out that in the studies cited, the decoupling global correlations that may be identified with first-order between the slab and the subslab mantle required by our contributions to the global subslab splitting signal. These model was not taken into account. Additionally, if the comparisons support a model in which the subslab flow subslab flow field is in fact dominated by entrained flow, field is controlled by the migration of trenches with respect as argued by Faccenda et al. [2008], then the anisotropy in to the convecting mantle, as previously proposed by Long

22 of 25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312 and Silver [2008]. Such a model requires a thin decoupling Buttles, J., and P. Olson (1998), A laboratory model of subduction zone anisotropy, Earth Planet. Sci. Lett., 164, 245–262, doi:10.1016/S0012- zone beneath slabs; we hypothesize that this is accom- 821X(98)00211-8. plished by entraining a thin layer of hot, buoyant astheno- Carlson, R. L. (1995), A plate cooling model relating rates of plate motion sphere, which in our model represents ambient mantle that to the age of the lithosphere at trenches, Geophys. Res. Lett., 22, 1977– 1980, doi:10.1029/95GL01807. has been subjected to shear heating during deformation, Cassidy, J. F., and M. G. Bostock (1996), Shear-wave splitting above the resulting in a viscosity reduction and temperature increase. subducting Juan de Fuca plate, Geophys. Res. Lett., 23,941–944, Our model also requires a partial barrier to flow, likely at the doi:10.1029/96GL00976. base of the transition zone, which permits downgoing slabs Civello, S., and L. Margheriti (2004), Toroidal mantle flow around the Calabrian slab (Italy) from SKS splitting, Geophys. Res. Lett., 31, to penetrate into the lower mantle but which prevents the L10601, doi:10.1029/2004GL019607. entrainment of the surrounding mantle material. Further Conrad, C. P., M. D. Behn, and P. G. Silver (2007), Global mantle flow and tests of our model for the creation and demise of oceanic the development of seismic anisotropy: Differences between the oceanic and continental upper mantle, J. Geophys. Res., 112, B07317, asthenosphere will include the investigation of splitting doi:10.1029/2006JB004608. parameters in regions where young lithosphere is being Cruciani, C., E. Carminati, and C. Doglioni (2005), Slab dip vs. lithosphere subducted and characterizations of the mantle temperature age: No direct function, Earth Planet. Sci. Lett., 238, 298–310, doi:10.1016/j.epsl.2005.07.025. beneath oceanic lithosphere from petrology and seismic Currie, C. A., J. F. Cassidy, R. Hyndman, and M. G. Bostock (2004), Shear constraints. wave anisotropy beneath the Cascadia subduction zone and western North American craton, Geophys. J. Int., 157, 341–353, doi:10.1111/ j.1365-246X.2004.02175.x. [64] Acknowledgments. We thank David Abt, Bill Holt, Chris Kincaid, Laurent Montesi, Jim Ni, Ray Russo, David Stegman, Wouter DeMets, C., R. G. Gordon, D. F. Argus, and S. Stein (1994), Effect of Schellart, and Jessica Warren for helpful and thought-provoking discus- recent revisions to the geomagnetic reversal time scale on estimates of sions. Wouter Schellart kindly provided us with his compilation of trench current plate motions, Geophys. Res. Lett., 21, 2191–2194, doi:10.1029/ migration velocities, and Megan Anderson shared results with us prior to 94GL02118. publication. We thank the authors of the many studies used in our global DiVenere, V., and D. V. Kent (1999), Are the Pacific and Indo-Atlantic compilation, without which this work would not be possible. We are hotspots fixed? Testing the plate circuit through Antarctica, Earth Planet. grateful to the thoughtful and constructive comments provided by two Sci. Lett., 170, 105–117, doi:10.1016/S0012-821X(99)00096-5. anonymous reviewers and by the Associate Editor, which helped to improve Dziewonski, A. M. (2005), The robust aspects of global seismic tomogra- the manuscript considerably. This work was supported by the Department phy, in Plates, Plumes, and Paradigms, edited by G. R. Foulger et al., of Terrestrial Magnetism, Carnegie Institution of Washington. Spec. Pap. Geol. Soc. Am., 338, 147–154. Faccenda, M., L. Burlini, T. V. Gerya, and D. Mainprice (2008), Fault- induced seismic anisotropy by hydration in subducting oceanic plates, Nature, 455, 1097–1100, doi:10.1038/nature07376. References Fischer, K. M., and D. A. Wiens (1996), The depth distribution of mantle Abt, D. L., K. M. Fischer, L. Martin, G. A. Abers, J. M. Protti, and anisotropy beneath the Tonga subduction zone, Earth Planet. Sci. Lett., W. Strauch (2005), Shear-wave splitting tomography in the Central 142, 253–260, doi:10.1016/0012-821X(96)00084-2. American subduction zone: Implications for flow and melt in the mantle Fischer, K. M., M. J. Fouch, D. A. Wiens, and M. S. Boettcher (1998), wedge, Eos Trans. AGU, 86(52), Fall Meet. Suppl., Abstract T31D-05. Anisotropy and flow in Pacific subduction zone back-arcs, Pure Appl. Abt, D. L., K. M. Fischer, G. A. Abers, W. Strauch, J. M. Protti, and Geophys., 151, 463–475, doi:10.1007/s000240050123. V. Gonza´lez (2009), Shear wave anisotropy beneath Nicaragua and Costa Fischer, K. M., E. M. Parmentier, A. R. Stine, and E. R. Wolf (2000), Rica: Implications for flow in the mantle wedge, Geochem. Geophys. Modeling anisotropy and plate-driven flow in the Tonga subduction Geosyst., 10, Q05S15, doi:10.1029/2009GC002375. zone back arc, J. Geophys. Res., 105, 16,181–16,191, doi:10.1029/ Anderson, M. L., G. Zandt, E. Triep, M. Fouch, and S. Beck (2004), 1999JB900441. Anisotropy and mantle flow in the Chile-Argentina subduction zone from Fjeldskaar, W. (1994), Viscosity and thickness of the asthenosphere de- shear wave splitting analysis, Geophys. Res. Lett., 31, L23608, tected from the Fennoscandian uplift, Earth Planet. Sci. Lett., 126, doi:10.1029/2004GL020906. 399–410, doi:10.1016/0012-821X(94)90120-1. Ando, M., Y. Ishikawa, and F. Yamazaki (1983), Shear wave polarization Fouch, M. J., and K. M. Fischer (1996), Mantle anisotropy beneath north- anisotropy in the upper mantle beneath Honshu, Japan, J. Geophys. Res., west Pacific subduction zones, J. Geophys. Res., 101, 15,987–16,002, 88, 5850–5864, doi:10.1029/JB088iB07p05850. doi:10.1029/96JB00881. Anglin, D. K., and M. J. Fouch (2005), Seismic anisotropy in the Izu-Bonin Fouch, M. J., and K. M. Fischer (1998), Shear wave anisotropy in the subduction system, Geophys. Res. Lett., 32, L09307, doi:10.1029/ Mariana subduction zone, Geophys. Res. Lett., 25, 1221–1224, 2005GL022714. doi:10.1029/98GL00650. Argus, D. F., and R. G. Gordon (1991), No-net-rotation model of current Fukao, Y. (1984), Evidence from core-reflected shear waves for anisotropy plate velocities incorporating plate motion model Nuvel-1, Geophys. Res. in the earth’s mantle, Nature, 309, 695–698, doi:10.1038/309695a0. Lett., 18, 2039–2042, doi:10.1029/91GL01532. Fukao, Y., S. Widiyantoro, and M. Obayashi (2001), Stagnant slabs in the Audoine, E., M. K. Savage, and K. Gledhill (2004), Anisotropic structure upper and lower mantle transition region, Rev. Geophys., 39, 291–323, under a back arc spreading region, the Taupo Volcanic Zone, New Zeal- doi:10.1029/1999RG000068. and, J. Geophys. Res., 109, B11305, doi:10.1029/2003JB002932. Funiciello, F., C. Faccenna, D. Giardini, and K. Regenauer-Lieb (2003), Baccheschi, P., L. Margheriti, and M. S. Steckler (2007), Seismic aniso- Dynamics of retreating slabs: 2. Insights from three-dimensional labora- tropy reveals focused mantle flow around the Calabrian slab (southern tory experiments, J. Geophys. Res., 108(B4), 2207, doi:10.1029/ Italy), Geophys. Res. Lett., 34, L05302, doi:10.1029/2006GL028899. 2001JB000896. Becker, T. W. (2008), Azimuthal seismic anisotropy constrains net rotation Gaboret, C., A. M. Forte, and J.-P. Montagner (2003), The unique dynamics of the lithosphere, Geophys. Res. Lett., 35, L05303, doi:10.1029/ of the Pacific Hemisphere mantle and its signature on seismic anisotropy, 2007GL032928. Earth Planet. Sci. Lett., 208, 219 – 233, doi:10.1016/S0012-821X(03) Behn, M. D., C. P. Conrad, and P. G. Silver (2004), Detection of upper 00037-2. mantle flow associated with the African superplume, Earth Planet. Sci. Garnero, E. J., and A. K. McNamara (2008), Structure and dynamics of Lett., 224, 259–274, doi:10.1016/j.epsl.2004.05.026. Earth’s lower mantle, Science, 320, 626–628, doi:10.1126/science. Bock, G., R. Kind, A. Rudloff, and G. Asch (1998), Shear wave anisotropy 1148028. in the upper mantle beneath the Nazca plate in northern Chile, J. Geo- Gledhill, K., and D. Gubbins (1996), SKS splitting and the seismic aniso- phys. Res., 103, 24,333–24,345, doi:10.1029/98JB01465. tropy of the mantle beneath the Hikurangi subduction zone, New Zealand, Bowman, J. R., and M. Ando (1987), Shear-wave splitting in the upper- Phys. Earth Planet. Inter., 95, 227–236, doi:10.1016/0031-9201(95) mantle wedge above the Tonga subduction zone, Geophys. J. R. Astron. 03118-9. Soc., 88, 25–41. Gordon, R. G., and D. M. Jurdy (1986), Cenozoic global plate motions, Brisbourne, A., G. Stuart, and J.-M. Kendall (1999), Anisotropic structure J. Geophys. Res., 91, 12,389–12,406, doi:10.1029/JB091iB12p12389. of the Hikurangi subduction zone, New Zealand–integrated interpreta- Greve, S. M., M. K. Savage, and S. D. Hofmann (2008), Strong variations tion of surface-wave and body-wave observations, Geophys. J. Int., 137, in seismic anisotropy across the Hikurangi subduction zone, North Island, 214–230, doi:10.1046/j.1365-246x.1999.00786.x.

23 of 25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312

New Zealand, Tectonophysics, 462, 7–21, doi:10.1016/j.tecto. Lallemand, S., A. Heuret, C. Faccenna, and F. Funiciello (2008), Subduc- 2007.07.011. tion dynamics as revealed by trench migration, Tectonics, 27, TC3014, Gripp, A. E., and R. G. Gordon (2002), Young tracks of hot spots and doi:10.1029/2007TC002212. current plate velocities, Geophys. J. Int., 150, 321–361, doi:10.1046/ Le´on Soto, G., J. F. Ni, S. P. Grand, E. Sandvol, R. W. Valenzuela, j.1365-246X.2002.01627.x. M. Guzma´n Speziale, J. M. Go´mez Gonza´lez, and T. Domı´nguez Reyes Gudmundsson, O., and M. Sambridge (1998), A regionalized upper mantle (2009), Mantle flow in the Rivera-Cocos subduction zone, Geophys. (RUM) seismic model, J. Geophys. Res., 103, 7121–7136, doi:10.1029/ J. Int., in press. 97JB02488. Lev, E., and B. H. Hager (2008), Rayleigh-Taylor instabilities with aniso- Hamilton, W. B. (2003), An alternative Earth, GSA Today, 13, 4–12, tropic lithospheric viscosity, Geophys. J. Int., 173, 806 –814, doi:10.1130/1052-5173(2003)013<0004:AAE>2.0.CO;2. doi:10.1111/j.1365-246X.2008.03731.x. Hammond, J. O., S. Kaneshima, J. Wookey, H. Inoue, T. Yamashina, and Levin, V., D. Droznin, J. Park, and E. Gordeev (2004), Detailed mapping of P. Harjadi (2006), Shear-wave splitting beneath Indonesia: Evidence for seismic anisotropy with local shear waves in southeastern Kamchatka, slab anisotropy, Eos Trans. AGU, 87(52), Fall Meet. Suppl., Abstract Geophys. J. Int., 158, 1009–1023, doi:10.1111/j.1365-246X.2004. MR23A-07. 02352.x. Healy, D., S. M. Reddy, N. E. Timms, E. M. Gray, and A. Vitale Brovarone Li, C., R. D. van der Hilst, E. R. Engdahl, and S. Burdick (2008), A new (2009), Trench-parallel fast axes of seismic anisotropy due to fluid-filled global model for P wave speed variations in Earth’s mantle, Geochem. cracks in subducting slabs, Earth Planet. Sci. Lett., 283, 75–86, Geophys. Geosyst., 9, Q05018, doi:10.1029/2007GC001806. doi:10.1016/j.epsl.2009.03.037. Long, M. D. (2009), Going with the mantle flow, Nat. Geosci., 2,10–11, Heuret, A., and S. Lallemand (2005), Plate motions, slab dynamics and doi:10.1038/ngeo398. back-arc deformation, Phys. Earth Planet. Inter., 149,31–51, Long, M. D., and P. G. Silver (2008), The subduction zone flow field from doi:10.1016/j.pepi.2004.08.022. seismic anisotropy: A global view, Science, 319, 315–318, doi:10.1126/ Hoernle, K., et al. (2008), Geochemical and geophysical evidence for arc- science.1150809. parallel flow in the mantle wedge beneath Costa Rica and Nicaragua, Long, M. D., and R. D. van der Hilst (2005), Upper mantle anisotropy Nature, 451, 1094–1098, doi:10.1038/nature06550. beneath Japan from shear wave splitting, Phys. Earth Planet. Inter., 151, Holtzman, B. K., D. L. Kohlstedt, M. E. Zimmerman, F. Heidelbach, 206–222, doi:10.1016/j.pepi.2005.03.003. T. Hiraga, and J. Hustoft (2003), Melt segregation and strain partitioning: Long, M. D., and R. D. van der Hilst (2006), Shear wave splitting from Implications for seismic anisotropy and mantle flow, Science, 301, 1227– local events beneath the Ryukyu arc: Trench-parallel anisotropy in the 1230, doi:10.1126/science.1087132. mantle wedge, Phys. Earth Planet. Inter., 155, 300–312, doi:10.1016/ Husson, L., C. P. Conrad, and C. Faccenna (2008), Tethyan closure, Andean j.pepi.2006.01.003. orogeny, and westward drift of the Pacific Basin, Earth Planet. Sci. Lett., Long, M. D., B. H. Hager, M. V. de Hoop, and R. D. van der Hilst (2007), 271, 303–310, doi:10.1016/j.epsl.2008.04.022. Two-dimensional modeling of subduction zone anisotropy and applica- Jarrard, R. D. (1986), Relations among subduction parameters, Rev. Geo- tion to southwestern Japan, Geophys. J. Int., 170, 839–856, doi:10.1111/ phys., 24, 217–284, doi:10.1029/RG024i002p00217. j.1365-246X.2007.03464.x. Jung, H., and S.-I. Karato (2001), Water-induced fabric transitions in oli- Marson-Pidgeon, K., and M. K. Savage (1997), Frequency-dependent ani- vine, Science, 293, 1460–1463, doi:10.1126/science.1062235. sotropy in Wellington, New Zealand, Geophys. Res. Lett., 24, 3297– Jung, H., Z. Jiang, I. Katayama, T. Hiraga, and S. Karato (2006), Effects of 3300, doi:10.1029/97GL03274. water and stress on the lattice preferred orientation in olivine, Tectono- Marson-Pidgeon, K., and M. K. Savage (2004), Shear-wave splitting varia- physics, 421, 1–22, doi:10.1016/j.tecto.2006.02.011. tions across an array in the lower North Island, New Zealand, Geophys. Jung, H., W. Mo, and H. W. Green (2009), Upper mantle seismic anisotropy Res. Lett., 31, L21602, doi:10.1029/2004GL021190. resulting from pressure-induced slip transition in olivine, Nat. Geosci., 2, Matcham, I., M. K. Savage, and K. R. Gledhill (2000), Distribution of 73–77, doi:10.1038/ngeo389. seismic anisotropy in the subduction zone beneath the Wellington, New Karato, S.-I., and H. Jung (1998), Water, partial melting and the origin of Zealand, Geophys. J. Int., 140, 1 –10, doi:10.1046/j.1365-246x.2000. the seismic low velocity and high attenuation zone in the upper mantle, 00928.x. Earth Planet. Sci. Lett., 157, 193– 207, doi:10.1016/S0012-821X(98) Mierdel, K., H. Keppler, J. R. Smyth, and F. Langenhorst (2007), Water 00034-X. solubility in aluminous orthopyroxene and the origin of Earth’s astheno- Karato, S.-I., and P. Wu (1993), Rheology of the upper mantle: A synthesis, sphere, Science, 315, 364–368, doi:10.1126/science.1135422. Science, 260, 771–778, doi:10.1126/science.260.5109.771. Morley, A. M., G. W. Stuart, J.-M. Kendall, and M. Reyners (2006), Mantle Karato, S.-I., H. Jung, I. Katayama, and P. Skemer (2008), Geodynamic wedge anisotropy in the Hikurangi subduction zone, central North Island, significance of seismic anisotropy of the upper mantle: New insights from New Zealand, Geophys. Res. Lett., 33, L05301, doi:10.1029/ laboratory studies, Annu.Rev.EarthPlanet.Sci., 36,59–95, 2005GL024569. doi:10.1146/annurev.earth.36.031207.124120. Mu¨ller, C. (2001), Upper mantle seismic anisotropy beneath Antarctica and Kawakatsu, H., P. Kumar, Y. Takei, M. Shinohara, T. Kanazawa, E. Araki, the Scotia Sea region, Geophys. J. Int., 147, 105 – 122, doi:10.1046/ and K. Suyehiro (2009), Seismic evidence for sharp lithosphere- j.1365-246X.2001.00517.x. asthenosphere boundaries of oceanic plates, Science, 324, 499 –502, Mu¨ller, C., B. Bayer, A. Eckstaller, and H. Miller (2008), Mantle flow in doi:10.1126/science.1169499. the South Sandwich subduction environment from source-side shear Kincaid, C., and R. W. Griffiths (2003), Laboratory models of the thermal wave splitting, Geophys. Res. Lett., 35, L03301, doi:10.1029/ evolution of the mantle during rollback subduction, Nature, 425, 58–62, 2007GL032411. doi:10.1038/nature01923. Nakajima, J., and A. Hasegawa (2004), Shear-wave polarization anisotropy Kirby, S. H., E. R. Engdahl, and R. P. Denlinger (1996), Intermediate-depth and subduction-induced flow in the mantle wedge of northern Japan, intraslab earthquakes and arc volcanism as physical expressions of crustal Earth Planet. Sci. Lett., 225, 365–377, doi:10.1016/j.epsl.2004.06.011. and uppermost mantle metamorphism in subducting slabs, in Subduction Nakajima, J., J. Shimizu, S. Hori, and A. Hasegawa (2006), Shear-wave Top to Bottom, Geophys. Monogr. Ser., vol. 96, edited by G. E. Bebout et splitting beneath the southwestern Kurile arc and northeastern Japan arc: al., pp. 195–214, AGU, Washington, D. C. A new insight into mantle return flow, Geophys. Res. Lett., 33, L05305, Kneller, E. A., and P. E. van Keken (2007), Trench-parallel flow and doi:10.1029/2005GL025053. seismic anisotropy in the Marianas and Andean subduction systems, Nettles, M., and A. M. Dziewonski (2008), Radially anisotropic shear Nature, 450, 1222–1225, doi:10.1038/nature06429. velocity structure of the upper mantle globally and beneath North Amer- Kneller, E. A., and P. E. van Keken (2008), Effect of three-dimensional slab ica, J. Geophys. Res., 113, B02303, doi:10.1029/2006JB004819. geometry on deformation in the mantle wedge: Implications for shear Ni, S., E. Tan, M. Gurnis, and D. Helmberger (2002), Sharp sides to the wave anisotropy, Geochem. Geophys. Geosyst., 9, Q01003, African superplume, Science, 296, 1850–1852, doi:10.1126/science. doi:10.1029/2007GC001677. 1070698. Kneller, E. A., M. D. Long, and P. E. van Keken (2008), Olivine fabric Nippress, S. E. J., N. J. Kusznir, and J.-M. Kendall (2007), LPO predicted transitions and shear-wave anisotropy in the Ryukyu subduction system, seismic anisotropy beneath a simple model of a mid-ocean ridge, Geo- Earth Planet. Sci. Lett., 268, 268–282, doi:10.1016/j.epsl.2008.01.004. phys. Res. Lett., 34, L14309, doi:10.1029/2006GL029040. Kreemer, C., W. E. Holt, and A. J. Haines (2003), An integrated global O’Neill, C., D. Mu¨ller, and B. Steinberger (2005), On the uncertainties in model of present-day plate motions and plate boundary deformation, hot spot reconstructions and the significance of moving hot spot reference Geophys. J. Int., 154, 8–34, doi:10.1046/j.1365-246X.2003.01917.x. frames, Geochem. Geophys. Geosyst., 6, Q04003, doi:10.1029/ Lallemand, S., A. Heuret, and D. Boutelier (2005), On the relationships 2004GC000784. between slab dip, back-arc stress, upper plate absolute motion, and crustal Park, J., and V. Levin (2002), Seismic anisotropy: Tracing plate dynamics nature in subduction zones, Geochem. Geophys. Geosyst., 6, Q09006, in the mantle, Science, 296, 485–489, doi:10.1126/science.1067319. doi:10.1029/2005GC000917.

24 of 25 B10312 LONG AND SILVER: SUBDUCTION ZONE MANTLE FLOW B10312

Peyton, V.,V.Levin, J. Park, M. Brandon, J. Lees, E. Gordeev, and A. Ozerov Stegman, D. R., J. Freeman, W. P. Schellart, L. Moresi, and D. May (2006), (2001), Mantle flow at a slab edge: Seismic anisotropy in the Kamchatka Influence of trench width on subduction hinge retreat rates in 3-D models region, Geophys. Res. Lett., 28, 379–382, doi:10.1029/2000GL012200. of slab rollback, Geochem. Geophys. Geosyst., 7, Q03012, doi:10.1029/ Phipps Morgan, J., W. J. Morgan, Y.-S. Zhang, and W. H. F. Smith (1995), 2005GC001056. Observational hints for a plume-fed sub-oceanic asthenosphere and its Steinberger, B., R. Sutherland, and R. J. O’Connell (2004), Prediction of role in mantle convection, J. Geophys. Res., 100, 12,753–12,768, Emperor-Hawaii seamount locations from a revised model of global doi:10.1029/95JB00041. motion and mantle flow, Nature, 430, 167–173, doi:10.1038/nature02660. Phipps Morgan, J., J. Hasenclever, M. Hort, L. Ru¨pke, and E. M. Parmentier Stern, R. J. (2002), Subduction zones, Rev. Geophys., 40(4), 1012, (2007), On subducting slab entrainment of buoyant asthenosphere, Terra doi:10.1029/2001RG000108. Nova, 19, 167–173, doi:10.1111/j.1365-3121.2007.00737.x. Stubailo, I., and P. Davis (2007), Shear wave splitting measurements and Pin˜ero-Feliciangeli, L., and J.-M. Kendall (2008), Sub-slab mantle flow interpretation beneath Acapulco-Tampico transect in Mexico, Eos Trans. parallel to the Caribbean plate boundaries: Inferences from SKS splitting, AGU, 88(52), Fall Meet. Suppl., Abstract T51B-0539. Tectonophysics, 462, 22–34, doi:10.1016/j.tecto.2008.01.022. Syracuse, E., and G. Abers (2006), Global compilation of variations in slab Polet, J., P. G. Silver, S. Beck, T. Wallace, G. Zandt, S. Ruppert, R. Kind, depth beneath arc volcanoes and implications, Geochem. Geophys. Geo- and A. Rudloff (2000), Shear wave anisotropy beneath the Andes from syst., 7, Q05017, doi:10.1029/2005GC001045. the BANJO, SEDA, and PISCO experiments, J. Geophys. Res., 105, Tackley, P. J. (2008), Geodynamics: Layer cake or plum pudding?, Nat. 6287–6304, doi:10.1029/1999JB900326. Geosci., 1, 157–158, doi:10.1038/ngeo134. Pozgay, S. H., D. A. Wiens, J. A. Conder, H. Shiobara, and H. Sugioka Tackley, P. J., S. Xie, T. Nakagawa, and J. W. Hernlund (2005), Numerical (2007), Complex mantle flow in the Mariana subduction system: Evi- and laboratory studies of mantle convection: Philosophy, accomplish- dence from shear wave splitting, Geophys. J. Int., 170, 371–386, ments, and thermochemical structure and evolution, in Earth’s Deep doi:10.1111/j.1365-246X.2007.03433.x. Mantle: Structure, Composition, and Evolution,editedbyR.D.van Rokosky, J. M., T. Lay, and E. J. Garnero (2006), Small-scale lateral varia- der Hilst et al., pp. 83–99, AGU, Washington, D. C. tions in azimuthally anisotropic D00 structure beneath the Cocos Plate, Tanimoto, T., and T. Lay (2000), Mantle dynamics and seismic tomography, Earth Planet. Sci. Lett., 248, 411–425, doi:10.1016/j.epsl.2006.06.005. Proc. Natl. Acad. Sci. U. S. A., 97, 12,409–12,410, doi:10.1073/pnas. Russo, R. M., and P. G. Silver (1994), Trench-parallel flow beneath the 210382197. Nazca plate from seismic anisotropy, Science, 263, 1105 – 1111, Tarduno, J., et al. (2003), The Emperor Seamounts: Southward motion of doi:10.1126/science.263.5150.1105. the Hawaiian hotspot plume in Earth’s mantle, Science, 301, 1064–1069, Rychert, C. A., and P. M. Shearer (2009), A global view of the lithosphere- doi:10.1126/science.1086442. asthenosphere boundary, Science, 324, 495– 498, doi:10.1126/science. Tarduno, J., H.-P. Bunge, N. Sleep, and U. Hansen (2009), The bent 1169754. Hawaiian-Emperor hotspot track: Inheriting the mantle wind, Science, Rychert, C. A., K. M. Fischer, and S. Rondenay (2005), A sharp litho- 324, 50–53, doi:10.1126/science.1161256. sphere-asthenosphere boundary imaged beneath eastern North America, Tovish, A., and G. Schubert (1978), Island arc curvature, velocity of con- Nature, 436, 542–545, doi:10.1038/nature03904. vergence and angle of subduction, Geophys. Res. Lett., 5, 329–332, Rydelek, P. A., and I. S. Sacks (1988), Asthenospheric viscosity inferred doi:10.1029/GL005i005p00329. from correlated land-sea earthquakes in north-east Japan, Nature, 336, Turcotte, D. L., and G. Schubert (1982), Geodynamics Applications of 234–237, doi:10.1038/336234a0. Continuum Physics to Geological Problems, John Wiley, New York. Sandvol, E., and J. Ni (1997), Deep azimuthal seismic anisotropy in the van der Hilst, R. D., S. Widiyantoro, and E. R. Engdahl (1997), Evidence southern Kurile and Japan subduction zones, J. Geophys. Res., 102, for deep mantle circulation from global tomography, Nature, 386, 578– 9911–9922, doi:10.1029/96JB03489. 584, doi:10.1038/386578a0. Savage, M. K. (1999), Seismic anisotropy and mantle deformation: What Volti, T., A. Gorbatov, H. Shiobara, H. Sugioka, K. Mochizuki, and have we learned from shear wave splitting?, Rev. Geophys., 37, 65–106, Y. Kaneda (2006), Shear-wave splitting in the Mariana trough—A rela- doi:10.1029/98RG02075. tion between back-arc spreading and mantle flow?, Earth Planet. Sci. Schellart, W. P., J. Freeman, D. R. Stegman, L. Moresi, and D. May (2007), Lett., 244, 566–575, doi:10.1016/j.epsl.2006.02.038. Evolution and diversity of subduction zones controlled by slab width, Wang, Y., and L. Wen (2007), Geometry and P and S velocity structure of Nature, 446, 308–311, doi:10.1038/nature05615. the ‘‘African Anomaly’’, J. Geophys. Res., 112, B05313, doi:10.1029/ Schellart, W. P., D. R. Stegman, and J. Freeman (2008), Global trench 2006JB004483. migration velocities and slab migration induced upper mantle volume Wessel, P., Y. Harada, and L. W. Kroenke (2006), Toward a self-consistent, fluxes: Constraints to find an Earth reference frame based on minimizing high-resolution absolute plate motion model for the Pacific, Geochem. viscous dissipation, Earth Sci. Rev., 88, 118–144, doi:10.1016/j.earscirev. Geophys. Geosyst., 7, Q03L12, doi:10.1029/2005GC001000. 2008.01.005. Wiens, D. A., J. A. Conder, and U. H. Faul (2008), The seismic structure Schmid, C., S. van der Lee, and D. Giardini (2004), Delay times and shear and dynamics of the mantle wedge, Annu. Rev. Earth Planet. Sci., 36, wave splitting in the Mediterranean region, Geophys. J. Int., 159, 275– 421–455, doi:10.1146/annurev.earth.33.092203.122633. 290, doi:10.1111/j.1365-246X.2004.02381.x. Wirth, E., and M. D. Long (2008), Frequency dependent shear wave split- Silver, P. G. (1996), Seismic anisotropy beneath the continents: Probing the ting beneath Japan and implications for the mantle wedge, Eos Trans. depths of geology, Annu. Rev. Earth Planet. Sci., 24, 385–432, AGU, 89(53), Fall Meet. Suppl., Abstract V31A-2115. doi:10.1146/annurev.earth.24.1.385. Wookey, J., J.-M. Kendall, and G. Ru¨mpker (2005), Lowermost mantle Silver, P. G., and M. K. Savage (1994), The interpretation of shear-wave anisotropy beneath the north Pacific from differential S-ScS splitting, splitting parameters in the presence of two anisotropic layers, Geophys. J. Geophys. J. Int., 161, 829–838, doi:10.1111/j.1365-246X.2005.02623.x. Int., 119, 949–963, doi:10.1111/j.1365-246X.1994.tb04027.x. Silver, P. G., R. W. Carlson, and P. Olson (1988), Deep slabs, geochemical heterogeneity and the large-scale structure of mantle convection: Inves- M. D. Long, Department of Geology and Geophysics, Yale University, tigation of an enduring paradox, Annu. Rev. Earth Planet. Sci., 16, 477– P.O. Box 208109, New Haven, CT 06518, USA. ([email protected]) 541, doi:10.1146/annurev.ea.16.050188.002401. Smith, G. P., D. A. Wiens, K. M. Fischer, L. M. Dorman, S. C. Webb, and J. A. Hildebrand (2001), A complex pattern of mantle flow in the Lau Backarc, Science, 292, 713–716, doi:10.1126/science.1058763.

25 of 25